Download Operator-valued measures, dilations, and the theory

Operator-Valued Measures, Dilations, and the
Theory of Frames
Deguang Han
David R. Larson
Bei Liu
Rui Liu
Department of Mathematics, University of Central Florida, Orlando, USA
E-mail address: [email protected]
Department of Mathematics, Texas A&M University, College Station, USA
E-mail address: [email protected]
Department of Mathematics, Tianjin University of Technology,
Tianjin 300384, P.R. China
E-mail address: [email protected]
Department of Mathematics and LPMC, Nankai University, Tianjin
300071, P.R. China
E-mail address: [email protected]
2010 Mathematics Subject Classification. Primary 46G10, 46L07, 46L10, 46L51,
47A20;
Secondary 42C15, 46B15, 46B25, 47B48
Key words and phrases. operator-valued measures, von Neumann algebras,
dilations, normal maps, completely bounded maps, frames.
Acknowledgements: The authors were all participants in the NSF funded
Workshop in Analysis and Probability at Texas A&M University. The first author
acknowledges partial support by a grant from the NSF. Bei Liu and Rui Liu
received partial support from the China Scholarship Council, the National Nature
Science Foundation of China, the Fundamental Research Fund for the Central
Universities of China, and the Tianjin Science & Technology Fund.
Abstract. We develop elements of a general dilation theory for operatorvalued measures. Hilbert space operator-valued measures are closely related
to bounded linear maps on abelian von Neumann algebras, and some of our
results include new dilation results for bounded linear maps that are not necessarily completely bounded, and from domain algebras that are not necessarily
abelian. In the non-cb case the dilation space often needs to be a Banach
space. We give applications to both the discrete and the continuous frame
theory. There are natural associations between the theory of frames (including continuous frames and framings), the theory of operator-valued measures
on sigma-algebras of sets, and the theory of continuous linear maps between
C ∗ -algebras. In this connection frame theory itself is identified with the special
case in which the domain algebra for the maps is an abelian von Neumann algebra and the map is normal (i.e. ultraweakly, or σ-weakly, or w*) continuous.
Some of the results for maps extend to the case where the domain algebra is
non-commutative. It has been known for a long time that a necessary and sufficient condition for a bounded linear map from a unital C*-algebra into B(H)
to have a Hilbert space dilation to a ∗-homomorphism is that the mapping
needs to be completely bounded. Our theory shows that even if it is not completely bounded it still has a Banach space dilation to a homomorphism. For
the special case when the domain algebra is an abelian von Neumann algebra
and the map is normal, we show that the dilation can be taken to be normal with respect to the usual Banach space version of ultraweak topology on
the range space. We view these results as generalizations of the known result
of Cazzaza, Han and Larson that arbitrary framings have Banach dilations,
and also the known result that completely bounded maps have Hilbertian dilations. Our methods extend to some cases where the domain algebra need
not be commutative, leading to new dilation results for maps of general von
Neumann algebras. This paper was motivated by some recent results in frame
theory and the observation that there is a close connection between the analysis of dual pairs of frames (both the discrete and the continuous theory) and
the theory of operator-valued measures.
Contents
Introduction
1
Chapter 1. Preliminaries
1.1. Frames
1.2. Operator-valued Measures
5
5
10
Chapter 2. Dilation of Operator-valued Measures
2.1. Basic Definitions
2.2. Dilation Spaces and Dilations
2.3. Elementary Dilation Spaces
2.4. The Minimal Dilation Norm k · kα
2.5. The Dual Space of the Minimal Dilation Space
2.6. The Maximal Dilation Norm k · kω
15
15
19
21
28
36
37
Chapter 3. Framings and Dilations
3.1. Hilbertian Dilations
3.2. Operator-Valued Measures Induced by Discrete Framings
3.3. Operator-Valued Measures Induced by Continuous Frames
43
43
46
50
Chapter 4. Dilations of Maps
4.1. Algebraic Dilations
4.2. The Commutative Case
4.3. The Noncommutative Case
57
57
58
64
Chapter 5.
71
Examples
Bibliography
81
v
Introduction
We investigate some natural associations between the theory of frames (including continuous frames and framings), the theory of operator-valued measures
(OVM’s) on sigma algebras of sets, and the theory of normal (ultraweakly or w*
continuous) linear maps on von Neumann algebras. Our main focus is on the dilation theory of these objects.
Generalized analysis-reconstruction schemes include dual pairs of frame sequences, framings, and continuous versions of these. We observe that all of these
induce operator-valued measures on an appropriate σ-algebra of Borel sets in a
natural way. The dilation theories for frames, dual pairs of frames, and framings,
have been studied in the literature and many of their properties are well known.
The continuous versions also have a dilation theory, but their properties are not as
well understood. We show that all these can be perhaps better understood in terms
of dilations of their operator-valued measures and their associated linear maps.
There is a well known dilation theory for those operator-valued measures that
are completely bounded in the sense that their associated bounded linear maps
between the operator algebra l∞ of the sigma algebra and the algebra of all bounded
linear operators on the underlying Hilbert space are completely bounded maps. In
this setting the dilation theory for the operator-valued measure (OVM) is obtained
naturally from the dilation theory for the cb map. In this theory, cb maps dilate
to *-homomorphisms while OVM’s dilate to projection valued measures (PVM’s),
where the projections are idempotent operators that are not-necessarily self-adjoint.
One of our main results shows that an OVM always has a dilation to a PVM even
if it is not completely bounded. This generalizes Naimark’s (Neumark’s) Dilation
Theorem for positive operator valued measures. But even in the case that the
underlying space is a Hilbert space the dilation space cannot always be taken to be
a Hilbert space. Thus elements of the theory of Banach spaces are essential in this
work. A key idea is the introduction of a certain norm, called the Minimal Dilation
Norm, on the space of bounded measurable functions on the measure space for an
OVM. Using this we show that every OVM has a minimal dilation to a PVM, where
minimality is meant up to Banach isomorphism of the dilation spaces.
One corollary of this result is a dilation theorem for not-necessarily-cb maps
with appropriate continuity properties from a commutative von Neumann algebra
into B(H). Motivated by some ideas in this construction we then prove a universal
dilation theorem for bounded linear mappings between Banach algebras. We prove
that this is a true generalization of our commutative theorem in an important
special case, and generalizes some of our results for maps of commutative von
Neumann algebras to the case where the von Neumann algebra is non-commutative.
These results are apparently new for mappings of von Neumann algebras. They
generalize special cases of Stinespring’s Dilation Theorem. The standard discrete
1
2
INTRODUCTION
Hilbert space frame theory is identified with the special case of our theory in which
the domain algebra is abelian and purely atomic, the map is completely bounded,
and the OVM is purely atomic and completely bounded with rank-1 atoms. For
discrete framings (c.f. [CHL]), which are the natural discrete generalization of
frames, the maps and measures are no longer always completely bounded and the
dilations no longer always be taken to be Hilbert space dilations.
At times the normality hypothesis mentioned in the first paragraph will be
weakened or dropped (see Theorem 4.10), or sometimes necessarily modified or
strengthened to achieve results (see Theorem 4.12 and Corollary 4.13), but on
balance normality seems to be the most natural condition to work with for the
linear maps associated with the connection between frame theory and operator
algebras.
As mentioned above, framings are the natural generalization of discrete frame
theory (more specifically: dual-frame pairs) to a non-Hilbertian setting, and even if
the underlying space is a Hilbert space the dilation space can fail to be Hilbertian.
This theory was originally developed by Casazza, Han and Larson in [CHL] as
an attempt to introduce frame theory with dilations into a Banach space context.
The initial motivation for the present manuscript was to completely understand
the dilation theory of framings. In the context of Hilbert spaces, we realized that
the dilation theory for discrete framings from [CHL] induces a dilation theory
for discrete operator valued measures that fail to be to be completely bounded
in the sense of (c.f. [Pa]). Our Theorem 3.8 states that a framing for a Hilbert
space is rescalable to a dual pair of frames if and only if the induced operator
valued measure has a Hilbert space dilation (equivalently the OVM is completely
bounded). Theorem 5.1 gives an example of a framing for a Hilbert space which is
not rescalable to a dual frame pair. The proof uses a non-completely bounded map
to construct a non-completely bounded OVM which yields the required framing.
This delimiting example shows that the dilation theory for framings in [CHL]
gives a true generalization of Naimarks’s Dilation Theorem for the discrete case.
This result led us to consider general (non-necessarily-discrete) operator valued
measures. Our main results of Chapter 2 give a dilation theory for general (nonnecessarily-cb) OVM’s that completely generalizes Naimark’s Dilation theorem in
a Banach space setting, and which is new even for Hilbert space.
There is also a well-known theory establishing a connection between general
bounded linear mappings from the C ∗ -algebra C(X) of continuous functions on a
compact Hausdorf space X into B(H) and operator valued measures on the sigma
algebra of Borel subsets of X (c.f. [Pa]). If A is an abelian C ∗ -algebra then A
can be identified with C(X) for a topological space X and can also be identified
with C(βX) where βX is the Stone-Cech compactification of X. Then the support
σ-algebra for the OVM is the sigma algebra of Borel subsets of βX which is enormous. However in our generalized (commutative) framing theory A will always
be an abelian von-Neumann algebra presented up front as L∞ (Ω, Σ, µ), with Ω a
topological space and Σ its algebra of Borel sets, and the maps on A into B(H)
are normal. In particular, to model the discrete frame and framing theory Ω is a
countable index set with the discrete topology (most often N), so Σ is its power set,
and µ is counting measure. So in this setting it is more natural to work directly
with this presentation in developing dilation theory rather than passing to βΩ, and
we take this approach in this paper. We feel that the connection we make with
INTRODUCTION
3
established discrete frame and framing theory is transparent, and then the OVM
dilation theory for the continuous case becomes a natural but nontrivial generalization of the theory for the discrete case that was inspired by framings. After doing
this we attempted to apply our techniques to the case where the domain algebra for
a map is non-commutative. We obtained some results which we present in Chapter
4. We show that all bounded maps have Banach space dilations. However, additional hypotheses are needed if dilations of maps are to have strong continuity
and structural properties. For a map between C*-algebras it is well-known that
there is a Hilbert space dilation if the map is completely bounded. (If the domain
algebra is commutative this statement is an iff.) Even if a map is not cb it has a
Banach space dilation. We are interested in the continuity and structural properties a dilation can have. In the discrete abelian case we show that the dilation of a
normal map can be taken to be normal and the dilation space can be taken to be
separable. Corollary 4.10 shows that with suitable hypotheses this type of result
can be generalized to the noncommutative setting.
This theory uses Hilbert space operator algebra theory and aspects of Banach
space theory, so we try to present Banach space versions of Hilbert space results
when we can obtain them. Some of the essential Hilbert space results we use are
proven more naturally in a wider Banach context. Chapter one contains preliminary
results and some exposition. In chapter two we develop our theory of Banach space
operator-valued measures and the accompanying dilation theory. Operator valued
measures have many different dilations to idempotent valued measures on larger
Banach spaces (even if the measure to be dilated is a cb measure on a Hilbert space)
and a part of the theory necessarily deals with classification issues. Chapter three is
devoted to some additional results and exposition for Hilbert space operator-valued
framings and measures, including the non-cb measures and their Banach dilations.
It contains exposition and examples on the manner in which frames and framings
on a Hilbert space induce natural operator valued measures on that Hilbert space.
The reader might well benefit by reading this chapter first, although doing that
would not be the natural order in which this theory is presented. In chapter 4 we
present our results on dilations of linear maps.
CHAPTER 1
Preliminaries
1.1. Frames
A frame F for a Hilbert space H is a sequence of vectors {xn } ⊂ H indexed by
a countable index set J for which there exist constants 0 < A ≤ B < ∞ such that,
for every x ∈ H,
X
(1.1)
Akxk2 ≤
| h x, xn i |2 ≤ Bkxk2 .
n∈J
The optimal constants (maximal for A and minimal for B) are known respectively
as the upper and lower frame bounds. A frame is called a tight frame if A = B,
and is called a Parseval frame if A = B = 1. If we only require that a sequence
{xn } satisfies the upper bound condition in (1.1), then {xn } is also called a Bessel
sequence.
A frame which is a basis is called a Riesz basis. Orthonormal bases are special
cases of Parseval frames. It is elementary that a Parseval frame {xn } for a Hilbert
space H is an orthonormal basis if and only if each xn is a unit vector.
For a Bessel sequence {xn }, its analysis operator Θ is a bounded linear operator
from H to `2 (N) defined by
X
(1.2)
Θx =
h x, xn i en ,
n∈N
where {en } is the standard orthonormal basis for `2 (N). It can be easily verified
that
Θ∗ en = xn , ∀n ∈ N
The Hilbert space adjoint Θ∗ is called the synthesis operator for {xn }. The positive
operator S := Θ∗ Θ : H → H is called the frame operator, or sometimes the Bessel
operator if the Bessel sequence is not a frame, and we have
X
(1.3)
Sx =
h x, xn i xn , ∀x ∈ H.
n∈N
A sequence {xn } is a frame for H if and only if its analysis operator Θ is
bounded, injective and has closed range, which is, in turn, equivalent to the condition that the frame operator S is bounded and invertible. In particular, {xn } is a
Parseval frame for H if and only if Θ is an isometry or equivalently if S = I.
Let S be the frame operator for a frame {xn }. Then the lower frame bound is
1/||S −1 || and the upper frame bound is ||S||. From (1.3) we obtain the reconstruction formula (or frame decomposition):
X
x=
x, S −1 xn xn , ∀x ∈ H
n∈N
5
6
1. PRELIMINARIES
or equivalently
x=
X
h x, xn i S −1 xn , ∀x ∈ H.
n∈N
(The second equation is obvious. The first can be obtained from the second by
replacing x with S −1 x and multiplying both sides by S.)
The frame {S −1 xn } is called the canonical or standard dual of {xn }. In the
case that {xn } is a Parseval frame for H, we have that S = I and hence
X
x=
h x, xn i xn , ∀x ∈ H.
n∈N
More generally, if a Bessel sequence {yn } satisfies
X
h x, yn i xn , ∀x ∈ H,
(1.4)
x=
n∈N
where the convergence is in norm of H, then {yn } is called an alternate dual of
{xn }. (Then {yn } is also necessarily a frame.) The canonical and alternate duals
are usually simply referred to as duals, and {xn , yn } is called a dual frame pair. It
is a well-known fact that that a frame {xn } is a Riesz basis if and only if {xn } has
a unique dual frame (cf. [HL]).
There is a geometric interpretation of Parseval frames and general frames. Let
P be an orthogonal projection from a Hilbert space K onto a closed subspace H, and
let {un } be a sequence in K. Then {P un } is called the orthogonal compression of
{un } under P , and correspondingly {un } is called an orthogonal dilation of {P un }.
We first observe that if {un } is a frame for K, then {P un } is a frame for H with
frame bounds at least as good as those of {un } (in the sense that the lower frame
cannot decrease and the upper bound cannot increase). In particular, {P un } is a
Parseval frame for H when {un } is an orthonormal basis for K, i.e., every orthogonal
compression of an orthonormal basis (resp. Riesz basis) is a Parseval frame (resp.
frame) for the projection subspace. The converse is also true: every frame can be
orthogonally dilated to a Riesz basis, and every Parseval frame can be dilated to an
orthonormal basis. This was apparently first shown explicitly by Han and Larson
in Chapter 1 of [HL]. There, with appropriate definitions it had an elementary
two-line proof. And as noted by several authors, it can be alternately derived by
applying the Naimark (Neumark) Dilation theorem for operator valued measures
by first passing from a frame sequence to a natural discrete positive operator-valued
measure on the power set of the index set. So it is sometimes referred to as the
Naimark dilation theorem for frames. In fact, this is the observation that inspired
much of the work in this paper.
For completeness we formally state this result:
Proposition 1.1. [HL] Let {xn } be a sequence in a Hilbert space H. Then
(i) {xn } is a Parseval frame for H if and only if there exists a Hilbert space
K ⊇ H and an orthonormal basis {un } for K such that xn = P un , where
P is the orthogonal projection from K onto H.
(ii) {xn } is a frame for H if and only if there exists a Hilbert space K ⊇ H
and a Riesz basis {vn } for K such that xn = P vn , where P again is the
orthogonal projection from K onto H.
The above dilation result was later generalized in [CHL] to dual frame pairs.
1.1. FRAMES
7
Theorem 1.2. Suppose that {xn } and {yn } are two frames for a Hilbert space
H. Then the following are equivalent:
(i) {yn } is a dual for {xn };
(ii) There exists a Hilbert space K ⊇ H and a Riesz basis {un } for K such
that xn = P un , and yn = P u∗n , where {u∗n } is the (unique) dual of the
Riesz basis {un } and P is the orthogonal projection from K onto H.
As in [CHL], a framing for a Banach space X is a pair of sequences {xi , yi }
with {xi } in X, {yi } in the dual space X ∗ of X, satisfying the condition that
X
x=
hx, yi ixi ,
i
where this series converges unconditionally for all x ∈ X.
The definition of a framing is a natural generalization of the definition of a
dual frame pair. Assume that {xi } is a frame for H and {yi } is a dual frame
for {xi }. Then {xi , yi } is clearly a framing for H. Moreover, if αi is a sequence
of non-zero constants, then {αi xi , ᾱi−1 yi } (called a rescaling of the pair) is also a
framing, although a simple example (Example 1.3) shows that it need not be a pair
of frames, even if {αi xi }, {ᾱi−1 yi } are bounded sequence.
Example 1.3. Let H be a separable Hilbert space and let {ei }i∈N be an orthonormal basis of H. Let
n
o
{xi } = e1 , e2 , e2 , e3 , e3 , e3 , . . .
and
Then {xi , yi }i∈N
are frames for H.
Proof. Let
o
n 1
1
1
1
1
{yi } = e1 , e2 , e2 , e3 , e3 , e3 , . . . .
2
2
3
3
3
is a framing of H, kxi k ≤ 1, kyi k ≤ 1, but neither {xi } nor {yi }
n 1
o
1
1
1
1
{αi } = 1, √ , √ , √ , √ , √ , . . . ,
2
2
3
3
3
then
n
o
1
1
1
1
1
{αi xi } = {yi /αi } = e1 , √ e2 , √ e2 , √ e3 , √ e3 , √ e3 , . . .
2
2
3
3
3
is a Parseval frame of H. Thus for any x ∈ H, we have
X
X
yi
x=
hx, αi xi i =
hx, xi iyi ,
αi
i∈N
i∈N
and this series converges unconditionally. Hence {xi , yi }i∈N is a framing of H and
kxi k ≤ 1, kyi k ≤ 1. But for any j ∈ N,
X
|hej , xi i|2 = jkej k = j
i∈N
and
X
i∈N
|hej , yi i|2 =
kej k
1
= .
j
j
So {xi } is not a Bessel sequence and {yi } is a Bessel sequence but not a frame.
8
1. PRELIMINARIES
If {xi , yi } is a framing, and if {xi , yi } are both Bessel sequences, then {xi }, {yi }
are frames and {xi , yi } is a dual frame pair. Indeed, if B, C are the Bessel bounds
for {xi }, {yi } respectively, then for each x,
*
+
X
hx, xi =
hx, yi ixi , x
i∈N
=
X
hx, yi ihxi , xi
i∈N
!1/2
≤
X
!1/2
X
2
|hx, yi i|
i∈N
2
|hxi , xi|
i∈N
!1/2
≤ C
1/2
kxk
X
2
|hxi , xi|
,
i∈N
P
2
and so C −1 kxk2 ≤ i∈N |hxi , xi| as required. So our interests will involve framings
for which {xi }, {yi }, or both, are not Bessel sequences.
Definition 1.4. [CHL] A sequence {xi }i∈N in a Banach space X is a projective
frame for X if there is a Banach space Z with an unconditional basis {zi , zi∗ } with
X ⊂ Z and a (onto) projection P : Z → X so that P zi = xi for all i ∈ N. If {zi } is
a 1-unconditional basis for Z and kP k = 1, we will call {xi } a Projective Parseval
frame for X.
In this case, we have for all x ∈ X that
X
X
X
x=
hx, zi∗ izi = P x =
hx, zi∗ iP zi =
hx, zi∗ ixi ,
i
i
i
and this series converges unconditionally in X. So this definition recaptures the
unconditional convergence from the Hilbert space definition.
We note that there exist projective frames in the sense of Definition 1.4 for an
infinite dimensional Hilbert space that fail to be frames. An example is contained
in Chapter 5.
Definition 1.5. [CHL] A framing model is a Banach space Z with a fixed
unconditional basis {ei } for Z. A framing modeled on (Z, {ei }i∈N ) for a Banach
space X is a pair of sequences {yi } in X ∗ . and {xi } in X so that the operator
θ : X → Z defined by
X
θu =
hu, yi iei ,
i∈N
is an into isomorphism and Γ : Z → X given by
X
X
Γ(
ai ei ) =
ai xi
i∈N
i∈N
is bounded and Γθ = IX .
In this setting, Γ becomes the reconstruction operator for the frame. The
following result due to Casazza, Han and Larson [CHL] shows that these three
methods for defining a frame on a Banach space are really the same.
Proposition 1.6. Let X be a Banach space and {xi } be a sequence of elements
of X. The following are equivalent:
1.1. FRAMES
9
(1) {xi } is a projective frame for X.
(2) There exists a sequence yi ∈ X ∗ so that {xi , yi } is a framing for X.
(3) There exists a sequence yi ∈ X ∗ and a framing model (Z, {ei }) so that
{xi , yi } is a framing modeled on (Z, {ei }).
This proposition tells us that if {xi , yi } is a framing of X, then {xi , yi } can be
dilated to an unconditional basis. That is, we can find a Banach space Z with an
unconditional basis {ei , e∗i }, X ⊂ Z and two bounded linear maps S and T such
that Sei = xi and T e∗i = yi .
Definition 1.7. Let H be a separable Hilbert space and Ω be a σ-locally
compact (σ-compact and locally compact) Hausdorff space endowed with a positive
Radon measure µ with supp(µ) = Ω. A weakly continuous function F : Ω → H is
called a continuous frame if there exist constants 0 < C1 ≤ C2 < ∞ such that
Z
2
C1 kxk ≤
|hx, F(ω)i|2 d µ(ω) ≤ C2 kxk2 , ∀ x ∈ H.
Ω
If C1 = C2 then the frame is called tight. Associated to F is the frame operator
SF defined in the weak sense by
Z
SF : H → H, hSF (x), yi := hx, F(ω)i · hF(ω), yid µ(ω).
Ω
It follows from the definition that SF is a bounded, positive, and invertible operator.
We define the following transform associated to F,
VF : H → L2 (Ω, µ),
VF (x)(ω) := hx, F(ω)i.
This operator is called the analysis operator in the literature and its adjoint operator
is given by
Z
∗
2
∗
VF : L (Ω, µ) → H, hVF (f ), xi :=
f (ω)hF(ω), xid µ(ω).
Ω
Then we have SF = VF∗ VF , and
Z
(1.5)
hx, yi = hx, F(ω)i · hG(ω), yid µ(ω),
Ω
−1
SF
F(ω)
where G(ω) :=
is the standard dual of F. A weakly continuous function
F : Ω → H is called Bessel if there exists a positive constant C such that
Z
|hx, F(ω)i|2 d µ(ω) ≤ Ckxk2 , ∀ x ∈ H.
Ω
It can be easily shown that if F : Ω → H is Bessel, then it is a frame for H if
and only if there exists a Bessel mapping G such that the reconstruction formula
(1.5) holds. This G may not be the standard dual of F. We will call (F, G) a dual
pair.
A discrete frame is a Riesz basis if and only if its analysis operator is surjective.
But for a continuous frame F, in general we don’t have VF (H) = L2 (Ω, µ). In
fact, this could happen only when µ is purely atomic. Therefore there is no Riesz
basis type dilation theory for continuous frames (however, we will see later that in
contrast the induced operator-valued measure does have projection valued measure
dilations). The following modified dilation theorem was due to Gabardo and Han
[GH]:
10
1. PRELIMINARIES
Theorem 1.8. Let F be a (Ω, µ)-frame for H and G be one of its duals. Suppose
that both VF (H) and VG (H) are contained in the range space M of the analysis
operator for some (Ω, µ)-frame. Then there is a Hilbert space K ⊃ H and a (Ω, µ)frame F̃ for K with P F̃ = F, P G̃ = G and VF̃ (H) = M, where G̃ is the standard
dual of F̃ and P is the orthogonal projection from K onto H.
Definition 1.9. Let X be a Banach space and Ω be a σ-locally compact
Hausdorff space. Let µ be a Borel measure on ω. A continuous framing on X is a
pair of maps (F , G ),
F : Ω → X, G : Ω → X ∗ ,
such that the equation
Z
E(F ,G ) (B)x, y =
hx, G (ω)ihF (ω), yidµ(ω)
B
for x ∈ X, y ∈ X ∗ , and B a Borel subset of Ω, defines an operator-valued probability
measure on Ω taking value in B(X) (see Definition 2.1). In particular, we require
the integral on the right to converge for each B ⊂ Ω. We have
Z
(1.6)
E(F ,G ) (B) =
F (ω) ⊗ G (ω)dE(ω)
B
where the integral converges in the sense of Bochner. In particular, since E(F ,G ) (Ω) =
IX , we have for any x ∈ X that
Z
hx, yi = hx, G (ω)ihF (ω), yidE(ω).
Ω
1.2. Operator-valued Measures
This section briefly discusses the well-known dilation theory for operator-valued
measures, and establishes the connections between framing dilations and dilations
of their associated operator-valued measures (more detailed discussion and investigation will be given in the subsequent chapters).
In operator theory, Naimark’s dilation theorem is a result that characterizes
positive operator-valued measures. Let Ω be a compact Hausdorff space,and let
B be the σ-algebra of all the Borel subsets of Ω. A B(H)-valued measure on Ω
is a mapping E : B → B(H) that is weakly countably additive, i.e., if {Bi } is a
countable collection of disjoint Borel sets with union B, then
X
hE(B)x, yi =
hE(Bi )x, yi
i
holds for all x, y in H. The measure is called bounded provided that
sup{kE(B)k : B ∈ B} < ∞,
and we let kϕk denote this supremum. The measure is called regular if for all x, y
in H, the complex measure given by
(1.7)
µx,y (B) = hE(B)x, yi
is regular.
Given a regular bounded B(H)-valued measure E, one obtains a bounded,
linear map
φE : C(Ω) → B(H)
1.2. OPERATOR-VALUED MEASURES
11
by
Z
hφE (f )x, yi =
(1.8)
f d µx,y .
Ω
Conversely, given a bounded, linear map φ : C(Ω) → B(H), if one defines regular
Borel measures {µx,y } for each x, y in H by the above formula (1.8), then for each
Borel set B, there exists a unique, bounded operator E(B), defined by formula
(1.7), and the map B → E(B) defines a bounded, regular B(H)-valued measure.
Thus, we see that there is a one-to-one correspondence between the bounded, linear
maps of C(Ω) into B(H) and the regular bounded B(H)-valued measures. Such
measures are called
(i) spectral if E(B1 ∩ B2 ) = E(B1 ) · E(B2 ),
(ii) positive if E(B) ≥ 0,
(iii) self-adjoint if E(B)∗ = E(B),
for all Borel sets B, B1 and B2 .
Note that if E is spectral and self-adjoint, then E(B) must be an orthogonal
projection for all B ∈ B, and hence E is positive.
Theorem 1.10 (Naimark). Let E be a regular, positive, B(H)-valued measure
on Ω. Then there exist a Hilbert space K, a bounded linear operator V : H → K,
and a regular, self-adjoint, spectral, B(K)-valued measure F on Ω, such that
E(B) = V ∗ F (B)V.
Stinespring’s dilation theorem is for completely positive maps on C ∗ -algebras.
Let A be a unital C ∗ - algebra. An operator-valued linear map φ : A → B(H) is
said to be positive if φ(a∗ a) ≥ 0 for every a ∈ A, and it is called completely positive
if for every n-tuple a1 , ..., an of elements in A, the matrix (φ(a∗i aj )) is positive in
the usual sense that for every n-tuple of vectors ξ1 , ..., ξn ∈ H, we have
n
X
(1.9)
hφ(ai a∗j )ξj , ξi i ≥ 0
i,j=1
or equivalently,
(φ(a∗i aj ))
is a positive operator on the Hilbert space H ⊗ Cn .
Theorem 1.11 (Stinespring’s dilation theorem). Let A be a unital C ∗ -algebra,
and let φ : A → B(H) be a completely positive map. Then there exists a Hilbert
space K, a unital ∗−homomorphism π : A → B(K), and a bounded operator V :
H → K with kφ(1)k = kV k2 such that
φ(a) = V ∗ π(a)V.
The following is also well known for commutative C ∗ -algebras:
Theorem 1.12 (cf. Theorem 3.11, [Pa]). Let B be a C ∗ -algebra, and let φ :
C(Ω) → B be positive. Then φ is completely positive.
This result together with Theorem 1.11 implies that Stinespring’s dilation theorem holds for positive maps when A is a unital commutative C ∗ -algebra.
A proof of Naimark dilation theorem by using Stinespring’s dilation theorem
can be sketched as follows: Let φ : A → B(H) be the natural extension of E to the
C ∗ -algebra A generated by all the characteristic functions of measurable subsets of
Ω. Then φ is positive, and hence is completely positive by Theorem 1.12. Apply
Stinespring’s dilation theorem to obtain a ∗−homomorphism π : A → B(K), and a
12
1. PRELIMINARIES
bounded, linear operator V : H → K such that φ(f ) = V ∗ π(f )V for all f in A. Let
F be the B(K)−valued measure corresponding to π. Then it can be verified that
F has the desired properties.
Let A be a C ∗ - algebra. We use Mn to denote the set of all n × n complex
matrices, and Mn (A) to denote the set of all n × n matrices with entries from A.
For the following theory see (c.f. [Pa]):
Given two C ∗ -algebras A and B and a map φ : A → B, obtain maps φn :
Mn (A) → Mn (B) via the formula
φn ((ai,j )) = (φ(ai,j )).
The map φ is called completely bounded if φ is bounded and kφkcb = supn kφn k is
finite.
Completely positive maps are completely bounded. In the other direction we
have Wittstock’s decomposition theorem [Pa]:
Proposition 1.13. Let A be a unital C ∗ -algebra, and let φ : A → B(H) be a
completely bounded map. Then φ is a linear combination of two completely positive
maps.
The following is a generalization of Stinespring’s representation theorem.
Theorem 1.14. Let A be a unital C ∗ -algebra, and let φ : A → B(H) be a
completely bounded map. Then there exists a Hilbert space K, a ∗−homomorphism
π : A → B(K), and bounded operators Vi : H → K, i = 1, 2, with kφkcb = kV1 k·kV2 k
such that
φ(a) = V1∗ π(a)V2
for all a ∈ A. Moreover, if kφkcb = 1, then V1 and V2 may be taken to be isometries.
Now let Ω be a compact Hausdorff space, let E be a bounded, regular, operatorvalued measure on Ω, and let φ : C(Ω) → B(H) be the bounded, linear map
associated with E by integration as described in section 1.4.1. So for any f ∈ C(Ω),
Z
hφ(f )x, yi =
f d µx,y ,
Ω
where
µx,y (B) = hE(B)x, yi
The OVM E is called completely bounded when φ is completely bounded. Using
Wittstock’s decomposition theorem, E is completely bounded if and only if it can
be expressed as a linear combination of positive operator-valued measures.
Let {xi }i∈J be a non-zero frame for a separable Hilbert space H. Let Σ be the
σ-algebra of all subsets of J. Define the mapping
X
E : Σ → B(H), E(B) =
xi ⊗ xi
i∈B
where x ⊗ y is the mapping on H defined by (x ⊗ y)(u) = hu, yix. Then E is a
regular, positive B(H)-valued measure. By Naimark’s dilation Theorem 1.10, there
exists a Hilbert space K, a bounded linear operator V : H → K, and a regular,
self-adjoint, spectral, B(K)-valued measure F on J, such that
E(B) = V ∗ F (B)V.
We will show (easily) that this Hilbert space K can be `2 , and the atoms xi ⊗ xi
of the measure dilates to rank-1 projections ei ⊗ ei , where {ei } is the standard
1.2. OPERATOR-VALUED MEASURES
13
orthonormal basis for `2 . That is K can be the same as the dilation space in
Proposition 1.1 (ii).
Similarly, suppose that {xi , yi }i∈J is a non-zero framing for a separable Hilbert
space H. Define the mapping
X
E : Σ → B(H), E(B) =
xi ⊗ yi ,
i∈B
for all B ∈ Σ. Then E is a B(H)-valued measure. We will show that this E also
has a dilation space Z. But this dilation space is not necessarily a Hilbert space, in
general, it is a Banach space and consistent with Proposition 1.6. The dilation is
essentially constructed using Proposition 1.6 (ii), where the dilation of the atoms
xi ⊗ yi corresponds to the projection ui ⊗ u∗i and {ui } is an unconditional basis for
the dilation space Z.
CHAPTER 2
Dilation of Operator-valued Measures
We develop a dilation theory of operator-valued measures to projection (i.e.
idempotent)-valued measures. Our main results show that this can be always
achieved for any operator-valued measure on a Banach space. Our approach is
to dilate an operator-valued measure to a linear space which is “minimal” in a certain sense and complete it in a norm compatable with the dilation geometry. Such
“dilation norms” are not unique, and we w:ill focus on two: one we call the minimal
dilation norm and one we call the maximal dilation norm. The applications of these
new dilation results to Hilbert space operator-valued measures will be discussed in
the subsequent chapters.
A positive operator-valued measure is a measure whose values are non-negative
operators on a Hilbert space. From Naimark’s dilation Theorem, we know that
every positive operator-valued measure can be dilated to a self-adjoint, spectral
operator-valued measure on a larger Hilbert space. But not all of the operatorvalued measures can have a Hilbert dilation space. For example, the operatorvalued measure induced by the framing in Chapter 5 does not have a Hilbert
dilation space. However it always admits a Banach dilation space, and moreover it can be dilated to a spectral operator-valued measure on a larger Banach
space. Thus, it is necessary to develop the dilation theory to include non-Hilbertian
operator-valued measures on a Hilbert space; i.e. B(H)-valued measures that have
no Hilbert dilation space. We obtain a similar result to Naimark’s dilation Theorem
for non-Hilbertian B(H)-valued measures. That is, for an arbitrary non-necessarilyHilbertian B(H)-valued measure E on (Ω, Σ), we show that there exists a Banach
space X, a spectral B(X)-valued measure F on (Ω, Σ), and bounded linear maps
S and T such that E(B) = SF (B)T holds for every B ∈ Σ.
2.1. Basic Definitions
Definition 2.1. Let X and Y be Banach spaces, and let (Ω, Σ) be a measurable
space. A B(X, Y )-valued measure on Ω is a map E : Σ → B(X, Y ) that is countably
additive in the weak operator topology; that is, if {Bi } is a disjoint countable
collection of members of Σ with union B, then
X
y ∗ (E(B)x) =
y ∗ (E(Bi )x)
i
∗
∗
for all x ∈ X and y ∈ Y .
We will use the symbol (Ω, Σ, E) if the range space is clear from context, or
(Ω, Σ, E, B(X, Y )), to denote this operator-valued measure system.
Remark 2.2. The Orlicz-Pettis theorem states that weak unconditional convergence and norm unconditional convergence of a series are the same in every
15
16
2. DILATION OF OPERATOR-VALUED MEASURES
P
Banach space (c.f.[DJT]). Thus wePhave that i E(Bi )x weakly unconditionally
converges to E(B)x if and only if i E(Bi )x strongly unconditionally converges
to E(B)x. So Definition 2.1 is equivalent to saying that E is strongly countably
additive, that is, if {Bi } is a disjoint countable collection of members of Σ with
union B, then
X
E(B)x =
E(Bi )x, ∀x ∈ X.
i
Definition 2.3. Let E be a B(X, Y )-valued measure on (Ω, Σ). Then the norm
of E is defined by
kEk = sup kE(B)k.
B∈Σ
We call E normalized if kEk = 1.
Remark 2.4. A B(X, Y )-valued measure E is always bounded, i.e.
sup kE(B)k < +∞.
(2.1)
B∈Σ
Indeed, for all x ∈ X and y ∗ ∈ Y ∗ , µx,y∗ (B) := y ∗ (E(B)x) is a complex measure on
(Ω, Σ). From complex measure theory (c.f.[Rud]), we know that µx,y∗ is bounded,
i.e.
sup |y ∗ (E(B)x)| < +∞.
B∈Σ
By the Uniform Boundedness Principle, we get (2.1).
Definition 2.5. A B(X)-valued measure E on (Ω, Σ) is called:
(i) an operator-valued probability measure if E(Ω) = IX ,
(ii) a projection-valued measure if E(B) is a projection on X for all B ∈ Σ,
(iii) a spectral operator-valued measure if for all A, B ∈ Σ, E(A ∩ B) = E(A) ·
E(B) (we will also use the term idempotent-valued measure to mean a
spectral-valued measure.)
With Definition 2.1, a B(X)-valued measure which is a projection-valued measure is always a spectral-valued measure (c.f. [Pa]). We give a proof for completeness. (Note that spectral operator-valued measures are clearly projection-valued
measures).
Lemma 2.6. Let E be a B(X)-valued measure. If for each B ∈ Σ, E(B) is a
projection (i.e. idempotent), then for any A, B ∈ Σ, E(A ∩ B) = E(A) · E(B).
Proof. If P and Q are projections on X, then P + Q is a projection on X if
and only if P Q = QP = 0.
For disjoint A, B ∈ Σ, we have E(A) + E(B) = E(A ∪ B) which is a projection.
From above, we obtain that E(A)·E(B) = E(A)·E(B) = 0. Then, for all A, B ∈ Σ,
we have
E(A) · E(B)
= E((A ∩ B c ) ∪ (A ∩ B)) · E((B ∩ Ac ) ∪ (B ∩ A))
= (E(A ∩ B c ) + E(A ∩ B)) · (E(B ∩ Ac ) + E(B ∩ A))
=
E(A ∩ B c ) · E(B ∩ Ac ) + E(A ∩ B c ) · E(B ∩ A) +
+E(A ∩ B) · E((B ∩ Ac ) + E(A ∩ B) · E(B ∩ A)
= E(A ∩ B) · E(B ∩ A) = E(A ∩ B).
2.1. BASIC DEFINITIONS
17
Definition 2.7. Let E : Σ → B(X, Y ) be an operator-valued measure. Then
the dual of E, denoted by E ∗ , is a mapping from Σ to B(Y ∗ , X ∗ ) which is defined
by E ∗ (B) = [E(B)]∗ for all B ∈ Σ.
Proposition 2.8. Let X and Y be Banach spaces. If X is reflexive, then
E ∗ : Σ → B(Y ∗ , X ∗ ) is an operator-valued measure.
Proof. Let {Bi } be a disjoint countable collection of members of Σ with union
B. Then for all x ∈ X and y ∗ ∈ Y ∗ , we have
X
X
(E ∗ (B)y ∗ )(x) = y ∗ (E(B)x) =
y ∗ (E(Bi )x) =
(E ∗ (Bi )y ∗ )(x).
i
i
Since X = X ∗∗ , this shows that E is weakly countably additive in the sense of
Definition 2.1.
Remark 2.9. The following example shows that there exists an operator-valued
measure E for which E ∗ is not weakly countably additive. So the above result is
sharp.
Example 2.10. Let {xi , yi }∞
i=1 ⊂ l1 × l∞ and 1 = (1, 1, 1, · · · ), where
{xi }∞
i=1 = {e1 , e1 , e1 , e2 , e3 , e4 , e5 , e6 , · · · }
and {ei }∞
i=1
∗ ∗ ∗ ∗ ∗ ∗
{yi }∞
i=1 = {1, -1, e1 , e2 , e3 , e4 , e5 , e6 , · · · },
is the standard unit vector basis. Let Ω = N, Σ = 2N and
X
E(B) =
xi ⊗ yi .
i∈B
Then it can be verified that E ∗ is not weakly countably additive.
Given two operator-valued measure space systems (Ω, Σ, E, B(X)) and (Ω, Σ, F, B(Y )).
We say that
(i) E and F are isometrically equivalent (or isometric) if there is a surjective
isometry U : X → Y such that E(B) = U −1 F (B)U for all B ∈ Σ.
(ii) E and F are similar (or isomorphic) if there is a bounded linear invertible
operator Q : X → Y such that E(B) = Q−1 F (B)Q for all B ∈ Σ.
The following property follows immediately from the definition:
Proposition 2.11. Let X and Y be Banach spaces and let E : Σ → B(X, Y )
be an operator-valued measure. Assume that W and Z are Banach spaces and that
T : W → X and S : Y → Z both are bounded linear operators. Then the mapping
F : Σ → B(W, Z) defined by
F (B) = SE(B)T, ∀ B ∈ Σ,
is an operator-valued measure.
Let X be a Banach space and {xi , yi }i∈N be a framing for X. Then as we have
introduced in Chapter 1 that the mapping E defined by
X
E : 2N → B(X), E(B) =
xi ⊗ yi ,
i∈B
is an operator-valued probability measure. We will call it the operator-valued probability measure induced by the framing {xi , yi }i∈N . The following lemma shows that
every operator-valued probability measure system (N, 2N , E, B(X)) with rank one
atoms is induced by a framing.
18
2. DILATION OF OPERATOR-VALUED MEASURES
Lemma 2.12. Let (N, 2N , E, B(X)) be an operator-valued probability measure
system with
rank(E({i})) ∈ {0, 1}
for all i ∈ N. Then there exists a framing {xi , yi }i∈N of X such that E is induced
by {xi , yi }i∈N . Moreover, if X is a Hilbert space and E is spectral, then the framing
{xi , yi }i∈N can be chosen as a pair of Riesz bases of X.
Proof. Without loss of generality, we can assume that for any i ∈ N, rank(E({i})) =
1. Then we can find xi ∈ X and yi ∈ X ∗ such that for any x ∈ X,
E({i})(x) = yi (x)xi = (xi ⊗ yi )(x).
It follows that
(2.2)
x = IX (x) = E(N)(x) =
X
i∈N
E({i})(x) =
X
(xi ⊗ yi )(x).
i∈N
By Definition 2.1, we know that the series in (2.2) converges unconditionally for all
x ∈ X. Hence {xi , yi }i∈N is a framing of X.
When X is a Hilbert space and E is spectral, since for all B ∈ 2N ,
X
E(B)x =
hx, yi ixi , ∀x ∈ X,
i∈B
we have for any i, j ∈ N and x ∈ X,
E({i})E({j})x = hx, yj ihxj , yi ixi .
When i = j, since
E({i})E({i})x = hx, yi ihxi , yi ixi = hx, yi ixi = E({i})x,
we get hxi , yi i = 1.
When i 6= j, we have
E({i})E({j})x = hx, yj ihxj , yi ixi = 0,
P
and so hxj , yi i = 0. If i ai xi = 0, then
!
X
X
E({j})
ai xi =
ai E({j})(xi ) = aj xj = 0,
i
i
and hence aj = 0. Thus {xi }i∈N is an unconditional basis. By Lemma 3.6.2 in [Ch],
we know that {xi /kxi k}i∈N is a Riesz basis of X, and hence {kxi kyi }i∈N is also a
Riesz basis of X. Clearly for all B ∈ 2N , we also have
X
E(B)x =
hx, kxi kyi ixi /kxi k, ∀x ∈ X.
i∈E
2.2. DILATION SPACES AND DILATIONS
19
2.2. Dilation Spaces and Dilations
Let (Ω, Σ) be a measurable space and E : Σ → B(X, Y ) be an operator-valued
measure.
Definition 2.13. A Banach space Z is called a dilation space of an operatorvalued measure space (Ω, Σ, E) if there exist bounded linear operators S : Z → Y
and T : X → Z, and a projection-valued measure space (Ω, Σ, F, B(Z)) such that
for any B ∈ Σ,
E(B) = SF (B)T.
We call S and T the corresponding analysis operator and synthesis operator, respectively, and use (Ω, Σ, F, B(Z), S, T ) to denote the corresponding dilation projectionvalued measure space system.
It is easy to see that the mappings
G1 : Σ → B(X, Z),
G1 (B) = F (B)T
G2 : Σ → B(X, Z),
G2 (B) = SF (B)
and
are both operator-valued measures. We call G1 and G2 the corresponding analysis
operator-valued measure and synthesis operator-valued measure, respectively.
Clearly, E = G2 · G1 . Indeed,
E(B) = SF (B)T = SF (B)F (B)T = G2 (B)G1 (B).
Remark 2.14. If X = Y and E(Ω) = IX , then S is a surjection, T is an
isomorphic embedding, and T S : Z → Z is a projection onto T (X).
Further, we have the following result:
Lemma 2.15. Let (Ω, Σ, E) be an operator-valued measure space and (Ω, Σ, F, B(Z), S, T )
be a corresponding dilation projection-valued measure space system.
(i) If E(Ω) : X → Y is an isomorphic operator, then S : Z → Y is a
surjection and T : X → Z is an isomorphic embedding.
(ii) If X = Y and E(Ω) = IX , then T SF (Ω), F (Ω)T S and F (Ω)T SF (Ω) are
all projections on Z.
Proof. (i) Since E(Ω) is an isomorphic operator, for any y ∈ Y, there exists
x ∈ X such that E(Ω)x = y. Hence (SF (Ω)T )(x) = E(Ω)(x) = y, and so S is
surjective.
Suppose that T x1 = T x2 for x1 , x2 ∈ X. Then we have (SF (Ω)T )(x1 ) =
(SF (Ω)T )(x2 ), and so E(Ω)x1 = E(Ω)x2 . Since E(Ω) is invertible, we get x1 = x2 .
Thus T is an isomorphic embedding.
(ii) When X = Y and E(Ω) = IX , we have
(T SF (Ω))2
= T SF (Ω)T SF (Ω) = T E(Ω)SF (Ω)
(F (Ω)T S)2
= F (Ω)T SF (Ω)T S = F (Ω)T IX S
= T IX SF (Ω) = T SF (Ω),
= F (Ω)T S,
20
2. DILATION OF OPERATOR-VALUED MEASURES
and
(F (Ω)T SF (Ω))2
= F (Ω)T SF (Ω)F (Ω)T SF (Ω)
= F (Ω)T SF (Ω)T SF (Ω)
= F (Ω)T IX SF (Ω)
= F (Ω)T SF (Ω).
Thus T SF (Ω), F (Ω)T S and F (Ω)T SF (Ω) are all projections on Z.
In general, the corresponding dilation projection-valued measure system is not
unique. However, we can always find a projection-valued probability measure system from a known dilation projection-valued measure system.
Proposition 2.16. Let (Ω, Σ, E) be an operator-valued measure space and let
(Ω, Σ, F, B(Z), S, T ) be a corresponding dilation projection-valued measure space
system. Define a mapping F̂ : Σ → B(F (Ω)Z) by
F̂ (B) = F (B)|F (Ω)Z ,
for all B ∈ Σ.
Then
Ŝ = S|F (Ω)Z : F (Ω)Z → Y
and
T̂ = F (Ω)T : X → F (Ω)Z
are both bounded linear operators. And F̂ is a spectral operator-valued probability
measure such that
E(B) = Ŝ F̂ (B)T̂ , B ∈ Σ.
Proof. Clearly Ŝ and T̂ are both well defined and are bounded linear operators. Since
F (B)F (Ω)Z = F (B)Z = F (Ω)F (B)Z ⊂ F (Ω)Z,
F̂ is well defined. It is clearly that F̂ is an operator-valued probability measure.
Moreover, for all A, B ∈ Σ,
F̂ (A)F̂ (B)
= F (A)|F (Ω)Z F (B)|F (Ω)Z
= F (A)F (B)|F (Ω)Z
= F (A ∩ B)|F (Ω)Z
= F̂ (A ∩ B).
It follows that F̂ is a spectral operator-valued measure. Finally, we have for any
B ∈ Σ and x ∈ X,
Ŝ F̂ (B)T̂ (x)
= S|F (Ω)Z F (B)|F (Ω)Z F (Ω)T (x)
= S|F (Ω)Z F (B)F (Ω)T (x)
= SF (B)T (x)
= E(B)(x).
Remark 2.17. It is easy to see that
Ŝ T̂ = Ŝ F̂ (Ω)T̂ = E(Ω).
2.3. ELEMENTARY DILATION SPACES
21
If X = Y and E(Ω) = IX , then
T̂ Ŝ = T̂ SF (Ω) = F (Ω)T Ŝ = F (Ω)T SF (Ω)
is a projection from F (Ω)Z onto T̂ Z = F (Ω)T (Z).
As a consequence of Proposition 2.16, we obtain:
Corollary 2.18. If an operator-valued measure can be dilated to a spectral
operator-valued measure, then it can be dilated to a spectral operator-valued probability measure.
2.3. Elementary Dilation Spaces
This section provides the first step in the construction of dilation spaces for
operator-valued measures. The main result (Theorem 2.26) shows that for any dilation projection-valued measure system (Ω, Σ, F, B(Z), S, T ) of an operator-valued
measure system (Ω, Σ, E, B(X, Y )), there exists an “elementary” dilation operatorvalued measure system which can be linearly isometrically embedded into (Ω, Σ, F, B(Z), S, T ).
Definition 2.19. Let Y be a Banach space and (Ω, Σ) be a measurable space.
A mapping ν : Σ → Y is called a vector-valued measure if ν is countably
additive; that is, if {Bi } is a disjoint countable collection of members of Σ with
union B, then
X
ν(B) =
ν(Bi ).
i
We use the notation (Ω, Σ, ν, Y ) for a vector-valued measure system.
Remark 2.20. By the Orlicz-Pettis Theorem, we know that ν countably additive is equivalent to ν weakly countably additive. That is
X
y ∗ (ν(B)) =
y ∗ (ν(Bi )),
i
for all y ∗ ∈ Y ∗ .
We use the symbol MYΣ to denote the linear space of all vector-valued measures
on (Ω, Σ) of Y .
Let X, Y be Banach spaces and (Ω, Σ, E, B(X, Y )) an operator-valued measure
system. For any B ∈ Σ and x ∈ X, define
EB,x : Σ → Y,
EB,x (A) = E(B ∩ A)x,
∀A ∈ Σ.
Then it is easy to see that EB,x is a vector-valued measure on (Ω, Σ) of Y and
EB,x ∈ MYΣ .
Let ME = span{EB,x : x ∈ X, B ∈ Σ}. Obviously, ME ⊂ MYΣ , and we will
refer it as the space induced by (Ω, Σ, E, B(X, Y )). We first introduce some linear
mappings on the spaces X, Y and ME .
Lemma 2.21. Let X, Y be Banach spaces and (Ω, Σ, E, B(X, Y )) an operatorN
N
valued measure system. For any {Ci }N
i=1 ⊂ C, {Bi }i=1 ⊂ Σ and {xi }i=1 ⊂ X, the
mappings
!
N
N
X
X
S : ME → Y, S
Ci EBi ,xi =
Ci E(Bi )xi
i=1
T : X → ME ,
i=1
T (x) = EΩ,x
22
2. DILATION OF OPERATOR-VALUED MEASURES
and
F (B) : ME → ME ,
F (B)
N
X
!
Ci EBi ,xi
=
i=1
N
X
Ci EB∩Bi ,xi ,
∀B ∈ Σ
i=1
are well-defined and linear.
N
Proof. Obviously T is well-defined. For any {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and
N
{Bi }i=1 ⊂ Σ, if
N
X
Ci EBi ,xi = 0,
i=1
then, by definition of Exi ,Bi , we have for any A ∈ Σ that
N
X
Ci EBi ,xi (A) =
i=1
N
X
Ci E(Bi ∩ A)xi = 0.
i=1
Let B = Ω. Then
N
X
Ci E(Bi )xi = 0.
i=1
Hence S is well-defined. Similarly we can verify that F (B) is also well-defined.
Clearly for any B ∈ Σ, S, T and F (B) are linear.
With the aid of Lemma 2.21, we can now give the definition of a dilation norm.
Note: In this manuscript we will in most cases use the traditional notation k · k
for a norm; however, in the case of dilation norms (especially) we will frequently
find it convenient to use the functional notation, typically D(·), for a norm, because
of the length of the expressions being normed. In this case we will sometimes also
write k · kD for this same norm when the meaning is clear, using the norming
function D to subscript the traditional norm notation.
Definition 2.22. Let ME be the space induced by (Ω, Σ, E, B(X, Y )). Let k · k
fE,k·k .
be a norm on ME . Denote this normed space by ME,k·k and its completion M
f
The norm on ME,k·k , with k · k := k · kD given by a norming function D as discussed
above, is called a dilation norm of Eif the following conditions are satisfied:
fE,D → Y defined on ME by
(i) The mapping SD : M
!
N
N
X
X
SD
Ci EBi ,xi =
Ci E(Bi )xi
i=1
i=1
is bounded.
fE,D defined by
(ii) The mapping TD : X → M
TD (x) = EΩ,x
is bounded.
fE,D ) defined by
(iii) The mapping FD : Σ → B(M
!
N
N
X
X
FD (B)
Ci EBi ,xi =
Ci EB∩Bi ,xi
i=1
i=1
is an operator-valued measure,
N
N
where {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and {Bi }i=1 ⊂ Σ.
2.3. ELEMENTARY DILATION SPACES
23
Theorem 2.23. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system.
fE,D is a dilation space
If a norm D is a dilation norm of E, then the Banach space M
f
of (Ω, Σ, E, B(X, Y )). Moreover, (Ω, Σ, FD , B(ME,D ), SD , TD ) is the corresponding
dilation projection-valued probability measure system.
Proof. For any x ∈ X and B1 , B2 , A ∈ Σ, we have
FD (B1 ∩ B2 )(EA,x ) = EB1 ∩B2 ∩A,x = FD (B1 )(EB2 ∩A,x ) = FD (B1 )FD (B2 )(EA,x ).
So FD is a spectral operator-valued measure.
For any B ∈ Σ, we get
SD FD (B)TD (x) = SD FD (B)(EΩ,x ) = SD (EB,x ) = E(B)(x),
∀x ∈ X,
fϕ,D ), SD , TD ) is the corand thus SD FD (E)TD = E(B). Therefore (Ω, Σ, FD , B(M
responding dilation projection-valued measure system.
Observe that for any x ∈ X and B ∈ Σ, we have FD (Ω)(EB,x ) = EB,x . Hence
FD (Ω) = IM
fϕ,D . The proof is complete.
Definition 2.24. Let ME be the space induced by (Ω, Σ, E, B(X, Y )) and
fE,D the elementary
D be a dilation norm of E. We call the Banach space M
f
dilation space of E and (Ω, Σ, FD , B(ME,D ), SD , TD ) the elementary dilation
operator-valued measure system.
In order to prove the main result (the existence of a dilation norm) in this
section, we need the following lemma:
Lemma 2.25. Let X and Y be Banach spaces, and let (Ω, Σ) be a measurable
space. Assume that X0 ⊂ X is dense in X. If a mapping E : Σ → B(X, Y ) is
strongly countably additive on X0 , i.e. if {Bi } is a disjoint countable collection of
members of Σ with union B then
X
(2.3)
E(B)x =
E(Bi )x,
∀x ∈ X0 ,
i
and if E is uniformly bounded on X0 , i.e. there exists a constant C > 0 such that
for any B ∈ Σ
kE(B)xk ≤ Ckxk, ∀x ∈ X0 ,
then E is an operator-valued measure.
Proof. Since X0 = X, E is uniformly bounded on X. For any N > 0, by (2.3),
if {Bi }N
i=1 is a disjoint collection of members of Σ, then
!
N
N
[
X
(2.4)
E
Bi x =
E(Bi )x, ∀x ∈ X0 .
i=1
S
N
i=1
Hence E
We have
Bi =
PN
i=1
i=1
E(Bi ) on X.
N
X
E(Bi ) = E
i=1
N
[
i=1
!
Bi ≤ C.
Let {Bi }∞
i=1 be a countable disjoint collection of elements of Σ with union B.
For any x ∈ X, there exists a sequence {xj }∞
j=1 ⊂ X0 such that
lim xj = x.
j→∞
24
2. DILATION OF OPERATOR-VALUED MEASURES
Observe that
N
X
E(Bi )x
E(B)x −
i=1
N
N
N
X
X
X
= E(B)x − E(B)xM + E(B)xM −
E(Bi )xM +
E(Bi )xM −
E(Bi )x
i=1
i=1
i=1
N
N
N
X
X
X
E(Bi )xM −
≤ kE(B)x − E(B)xM k + E(B)xM −
E(Bi )xM + E(Bi )x
i=1
i=1
i=1
!
N
N
[
X
Bi · kxM − xk
≤ kE(B)k · kx − xM k + E(B)xM −
E(Bi )xM + E
i=1
i=1
N
X
≤ 2C kx − xM k + E(B)xM −
E(Bi )xM .
i=1
For any ε > 0, we can find M > 0 such that
ε
kx − xM k ≤
.
3C
By (2.3), for xM ∈ X0 , we can find a sufficiently large N > 0 such that
N
ε
X
E(Bi )xM ≤ .
E(B)xM −
3
i=1
Therefore
N
X
E(Bi )x ≤ ε
E(B)x −
i=1
when N is sufficiently large. Hence
X
E(B)x =
E(Bi )x,
∀x ∈ X.
i
Therefore E is an operator-valued measure.
Now we state and prove the main result of this section.
Theorem 2.26. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system and (Ω, Σ, F, B(Z), S, T ) be a corresponding dilation projection-valued measure
space system. Then there exist an elementary dilation operator-valued measure sysfE,D ), SD , TD ) of E and a linear isometric embedding
tem (Ω, Σ, FD , B(M
fE,D → Z
U :M
such that
SD = SU, F (Ω)T = U TD , U FD (B) = F (B)U,
∀B ∈ Σ.
Proof. Define D : ME → R+ ∪ {0} by
! N
N
X
X
D
Ci EBi ,xi = Ci F (Bi )T (xi ) ,
i=1
i=1
Z
2.3. ELEMENTARY DILATION SPACES
25
N
where N > 0, Ci are all complex numbers and {xi }N
i=1 ⊂ X and {Bi }i=1 ⊂ Σ. We
first show that D is a norm on ME .
(i) Obviously,
! N
N
X
X
D
Ci EBi ,xi = Ci F (Bi )T (xi ) ≥ 0.
i=1
i=1
Z
ej }M , {yj }M ⊂ X and {Aj }M ⊂ Σ. We have
(ii) Take {C
j=1
j=1
j=1


N
N
M
M
X
X
X
X
ej EA ,y  = e
D
Ci EBi ,xi +
C
F
(B
)T
(x
)
+
C
C
F
(A
)T
(y
)
i
i
i
j
j
j j j
i=1
i=1
j=1
j=1
Z
N
M
X
X
e
Cj F (Aj )T (yj )
Ci F (Bi )T (xi ) + ≤ j=1
i=1
Z
Z


!
N
M
X
X
ej EA ,y  .
C
= D
Ci EBi ,xi + D 
j j
i=1
(iii) If D
PN
i=1
P
N
i=1
Ci EBi ,xi
j=1
P
N
= 0, then i=1 Ci F (Bi )T (xi ) = 0, and hence
Z
Ci F (Bi )T (xi ) = 0. This implies that for any A ∈ Σ, we have
N
X
Ci EBi ,xi (A)
=
i=1
N
X
Ci E(Bi ∩ A)(xi ) =
i=1
N
X
Ci SF (Bi ∩ A)T (xi )
i=1
= SF (A)
N
X
Ci F (Bi )T (xi ) = 0.
i=1
Thus D is a norm on ME .
fE,D be the completion of ME,D under
Denote this norm by k · kD , and let M
this norm. We will show that D is a dilation norm of E.
First, since
N
!
N
X
X
Ci E(Bi )xi Ci EBi ,xi = SD
i=1
i=1
Y
N
Y !
N
X
X
= Ci SF (Bi )T xi = S
Ci F (Bi )T xi i=1
i=1
Y
N
N
Y
X
X
≤ kSk · Ci F (Bi )T xi = kSk · Ci EBi ,xi ,
i=1
i=1
Z
D
and
kTD (x)kD
= kEΩ,x kD = kF (Ω)T (x)kZ
≤
kF (Ω)k · kT k · kxkX ,
we have that the mappings
fE,D → Y,
SD : M
SD
N
X
i=1
!
Ci EBi ,xi
=
N
X
i=1
Ci E(Bi )xi
26
2. DILATION OF OPERATOR-VALUED MEASURES
and
fE,D ,
TD : X → M
TD (x) = EΩ,x
are both well-defined, linear and bounded.
fE,D ) defined by
We prove that the mapping FD : Σ → B(M
!
N
N
X
X
FD (B)
Ci EBi ,xi =
Ci EB∩Bi ,xi ,
i=1
i=1
is an operator-valued measure. By Lemma 2.25, we only need to show that FD is
strongly countably additive and uniformly bounded on ME,D .
If {Aj }∞
j=1 is a disjoint countable collection of members of Σ with union A,
since F is an operator-valued measure, we have
F (A ∩ Bi )T (xi ) =
∞
X
F (Aj ∩ Bi )T (xi ).
j=1
It follows that
FD (A)(EBi ,xi ) =
∞
X
FD (Aj )(EBi ,xi ).
j=1
Hence
FD (A)
N
X
!
Ci EBi ,xi
=
i=1
∞
X
FD (Aj )
N
X
j=1
!
Ci EBi ,xi
.
i=1
Thus FD is strongly countably additive on ME,D .
For any A ∈ Σ, we have
!
N
N
X
X
Ci EBi ,xi = Ci FD (A) (EBi ,xi )
FD (A)
i=1
i=1
D
D
N
N
X
X
= Ci EA∩Bi ,xi , = Ci F (A ∩ Bi )T (xi )
i=1
i=1
D
Z
N
N
X
X
= Ci F (A)F (Bi )T (xi ) ≤ kF (A)k Ci F (Bi )T (xi )
i=1
i=1
Z
Z
N
X
≤ sup kF (A)k Ci EBi ,xi .
A∈Σ
i=1
D
By Remark 2.4, we know that
sup kF (A)k < +∞,
A∈Σ
and thus FD is uniformly bounded on ME,D .
fE,D → Z by
Define a mapping U : M
!
N
N
X
X
U
Ci EBi ,xi =
Ci F (Bi )T (xi ).
i=1
i=1
It is easy to see that U is a well-defined and linear isometric embedding mapping.
2.3. ELEMENTARY DILATION SPACES
27
Finally, we have
N
X
SD
!
Ci EBi ,xi
=
i=1
N
X
Ci E(Bi )xi =
i=1
=
S
N
X
N
X
Ci SF (Bi )T (xi )
i=1
N
X
Ci F (Bi )T (xi ) = SU
i=1
!
Ci EBi ,xi
,
i=1
U TD (x) = U (EΩ,x ) = F (Ω)T (x)
and
U FD (B)
N
X
!
Ci EBi ,xi
= U
i=1
N
X
!
Ci EB∩Bi ,xi
=
i=1
=
N
X
N
X
Ci F (B ∩ Bi )T (xi )
i=1
Ci F (B)F (Bi )T (xi ) = F (B)
!
Ci F (Bi )T (xi )
i=1
i=1
= F (B)U
N
X
N
X
!
Ci EBi ,xi
.
i=1
Thus we get
SD = SU, F (Ω)T = U TD , U FD (B) = F (B)U,
∀B ∈ Σ,
as claimed.
In general, we can find different dilation norms on ME , and hence E may have
different dilation projection-valued measure systems. The following result shows
that if E is a projection B(X)-valued probability measure, then these different
dilation projection-valued measure systems are all similar, that is, E is isomorphic
to FD for every dilation norm D on ME .
Proposition 2.27. Let (Ω, Σ, E, B(X)) be a projection-valued probability
measure system and D be a dilation norm of E. Then SD and TD both are invertible operators with SD TD = IX and TD SD = IM
fE,D . Thus, E is isomorphic to
FD for every dilation norm D of E.
fE,D ), SD , TD ) is a
Proof. By Theorem 2.23, we know that (Ω, Σ, FD , B(M
projection-valued probability measure space system. Hence FD (Ω) = IM
fE,D and
SD TD = SD FD (Ω)TD = E(Ω) = IX .
By Remark 2.14, we know that SD is a surjection and TD is an isomorphic embedding.
N
N
Let N > 0, {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and {Bi }i=1 ⊂ Σ. If
!
N
N
X
X
SD
Ci EBi ,xi =
Ci E(Bi )xi = 0,
i=1
i=1
then for any A ∈ Σ,
N
X
Ci EBi ,xi (A) =
i=1
So we get
N
X
i=1
PN
i=1
Ci E(Bi ∩ A)(xi ) = E(A)
N
X
!
Ci E(Bi )xi
i=1
Ci EBi ,xi = 0, and hence SD is invertible.
= 0.
28
2. DILATION OF OPERATOR-VALUED MEASURES
Observe that for any A ∈ Σ,
EΩ,PN
(A)
i=1 Ci E(Bi )xi
= E(A)
N
X
!
Ci E(Bi )xi
=
i=1
N
X
Ci EBi ,xi (A).
i=1
Thus we have
TD SD
N
X
!
Ci EBi ,xi
= TD
i=1
=
N
X
!
Ci E(Bi )xi
i=1
EΩ,PN
=
i=1 Ci E(Bi )xi
N
X
Ci EBi ,xi .
i=1
So we get TD SD = IM
fϕ,D and TD is invertible. Therefore E is isomorphic to FD for
every dilation norm D of E.
2.4. The Minimal Dilation Norm k · kα
While the connection between a corresponding dilation projection-valued measure system and the elementary dilation space has been established in the previous
section, we still need to address the existence issue for a corresponding dilation
projection-valued measure system. So in this and the next two sections we will
focus on constructing two special dilation norms for the (algebraic) elementary
dilation space ME which we call the minimal and maximal dilation norms.
Define k · kα : ME → R+ ∪ {0} by
N
N
X
X
Ci E(B ∩ Bi )xi .
Ci EBi ,xi = sup B∈Σ
i=1
α
i=1
Y
Proposition 2.28. k · kα is a norm on ME .
N
Proof. Let N > 0, {Ci }N
⊂ C, {xi }N
i=1 ⊂ X and {Bi }i=1 ⊂ Σ. If
PN i=1
0, then for any B ∈ Σ, i=1 Ci E(B ∩ Bi )xi = 0, hence
PN
i=1
Ci EBi ,xi =
N
N
X
X
Ci EBi ,xi = sup Ci E(B ∩ Bi )xi = 0.
B∈Σ i=1
i=1
α
Y
Thus k· kα is well-defined.
PN
PN
If i=1 Ci EBi ,xi = 0, then for any B ∈ Σ, i=1 Ci E(B ∩ Bi )xi = 0, thus
α
N
X
i=1
Ci EBi ,xi = 0.
2.4. THE MINIMAL DILATION NORM k · kα
29
ej }M ⊂
Now we prove that k·kα satisfies the triangle inequality. Let M > 0, {C
j=1
M
M
C, {yj }j=1 ⊂ X and {Aj }j=1 ⊂ Σ. Then
N
M
X
X
e
Ci EBi ,xi +
Cj EAj ,yj i=1
j=1
α
N
M
X
X
e
= sup Ci E(Bi ∩ B)xi +
Cj E(Aj ∩ B)yj B∈Σ i=1
j=1
Y
N
M
X
X
e
Cj E(Aj ∩ B)yj ≤ sup Ci E(Bi ∩ B)xi + sup B∈Σ j=1
B∈Σ i=1
Y
Y
N
M
X
X
Cj EAj ,yj = Ci EBi ,xi + .
j=1
i=1
α
α
So k · kα is a norm on ME .
fE,α .
We denote its normed space by ME,α and its completion by M
Theorem 2.29. k · kα is a dilation norm of E.
N
N
Proof. Let N > 0, {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and {Bi }i=1 ⊂ Σ. Since the
f
linear map Sα : ME,α → Y is defined by
!
N
N
X
X
Ci E(Bi ∩ B)xi ,
Sα
Ci EBi ,xi =
i=1
i=1
we have that
N
!
X
Ci E(Bi ∩ B)xi Ci EBi ,xi = i=1
i=1
Y
Y
N
N
X
X
Ci EBi ,xi ,
Ci E(B ∩ Bi )xi = ≤ sup B∈Σ Sα
N
X
i=1
Y
i=1
M
which implies that Sα is bounded and kSα k ≤ 1.
fE,α is defined by Tα (x) = EΩ,x . Obviously,
Recall that the map Tα : X → M
we have
kTα xkα = kEΩ,x kα = sup kE(B)xkY ≤ kEk · kxk,
B∈Σ
and hence kTα k ≤ kEk.
fE,α ) defined by
Now we prove the mapping Fα : Σ → B(M
!
N
N
X
X
Fα (B)
Ci EBi ,xi =
Ci EB∩Bi ,xi
i=1
i=1
is an operator-valued measure. By Lemma 2.25, we only need to prove Fα is strongly
countably additive and uniform bounded on ME,α .
30
2. DILATION OF OPERATOR-VALUED MEASURES
Since
=
=
N
!
N
X
X
Ci EB∩Bi ,xi Ci EBi ,xi = Fα (B)
i=1
i=1
α
N α
N
X
X
Ci E(Bi ∩ B 0 )xi sup Ci E(Bi ∩ B ∩ B 0 )(xi ) = sup 0
0
B ∈Σ i=1
B ∈Σ i=1
Y
Y
N
X
Ci EBi ,xi ,
i=1
Y
we get that kFα (B)k = 1 for any B ∈ Σ.
For the strong countable additivity, let {Aj }∞
j=1 be a countable disjoint collection of members of Σ with union A. Then we have
!
!
X
N
N
X
X
M
Fα (Aj )
Ci EBi ,xi − Fα (A)
Ci EBi ,xi j=1
i=1
i=1
α
!
!
X
M
N
N
X
X
= Ci EAj ∩Ai ,xi −
Ci EA∩Bi ,xi j=1 i=1
i=1
α


N
M
X
X
Ci 
EAj ∩Bi ,xi − EA∩Bi ,xi 
= i=1
j=1
α


N
M
X
X
0
0
Ci 
E(Aj ∩ Bi ∩ B )xi − E(A ∩ Bi ∩ B )xi 
= sup B 0 ∈Σ i=1
j=1
Y

 N
∞
[
X
Ci E 
(Aj ∩ Bi ∩ B 0 ) xi = sup B 0 ∈Σ i=1
j=M +1

 Y
N
∞
X
[
0 

≤
|Ci | sup B
(A
∩
B
∩
B
)
x
j
i
i
B 0 ∈Σ i=1
j=M +1
Y

 N
∞
X
[
=
|Ci | sup E
(Aj ∩ B 0 ) xi B 0 ∈Σ i=1
j=M +1
Y
∞
N
X
X
=
|Ci | sup E(Aj ∩ B 0 )xi .
B 0 ∈Σ i=1
j=M +1
Y
P
∞
If supB 0 ∈Σ j=M +1 E(Aj ∩ B 0 )xi Y
does not tend to 0 as M → ∞, then we can
find δ > 0, a sequence of n1 ≤ m1 < n2 ≤ m2 < n3 ≤ m3 < . . . , and {Bl0 }∞
l=1 ⊂ Σ
such that
X
ml
0
E(Aj ∩ Bl )xi ≥ δ, ∀ l ∈ N.
j=nl
2.4. THE MINIMAL DILATION NORM k · kα
31
Since for l ∈ N and nl ≤ j ≤ ml , Aj ∩ Bl0 are disjoint from each other, we have


ml
ml
∞ [
∞ X
[
X
E
Aj ∩ Bl0  xi =
E(Bj ∩ Bl0 )xi ,
l=1 j=nl
l=1 j=nl
P
mi
0
E(B
∩
B
)x
which implies j=n
→ 0, which is a contradiction. Hence
j
i
i
i
!
!
N
∞
N
X
X
X
Fα (A)
Ci EBi ,xi =
Fα (Aj )
Ci EBi ,xi ,
i=1
j=1
i=1
as expected.
Combining Theorem 2.29 and Theorem 2.23 we have the following:
Theorem 2.30. For any operator-valued measure system (Ω, Σ, E, B(X, Y )),
there exist a Banach space Z, two bounded linear operators S : Z → Y and T :
X → Z, and a projection-valued probability measure system (Ω, Σ, F, B(Z)) such
that E(B) = SF (B)T for any B ∈ Σ. In other words, every operator-valued
measure can be dilated to a projection-valued measure.
Lemma 2.31. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system.
If D is a dilation norm of E, then there exists a constant CD such that for any
PN
i=1 Ci EBi ,xi ∈ ME,D ,
N
N
X
X
sup Ci E(B ∩ Bi )xi ≤ CD Ci EBi ,xi ,
B∈Σ
i=1
where N > 0,
that
{Ci }N
i=1
i=1
Y
⊂ C,
{xi }N
i=1
⊂ X and
kf kα ≤ CD kf kD ,
{Bi }N
i=1
D
⊂ Σ. Consequently we have
∀f ∈ ME .
Proof. Let CD = supB∈Σ kSD FD (B)k . Then obviously CD < +∞. We also
have
N
N
X
X
sup Ci E(B ∩ Bi )xi = sup Ci SD FD (B ∩ Bi )TD (xi )
B∈Σ i=1
B∈Σ i=1
Y
Y
N
X
= sup Ci SD FD (B)FD (Bi )TD (xi )
B∈Σ i=1
Y
N
X
≤ sup kSD FD (B)k Ci FD (Bi )TD (xi )
B∈Σ
i=1
D
N
X
= CD Ci EBi ,xi .
i=1
D
This lemma justifies the following definition for the dilation norm α.
Definition 2.32. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure sysfE,α the α-dilation space.
tem. We call α the minimal dilation norm of E and M
32
2. DILATION OF OPERATOR-VALUED MEASURES
The next result shows that if we start with a projection-valued probability measure E, then the dilated projection-valued probability measure on the elementary
fE,α is isometric to E.
dilation space M
Proposition 2.33. Let (Ω, Σ, E, B(X)) be a normalized projection-valued
probability measure system and α be the minimal dilation norm of E. Then Sα
and Tα both are isometries and E is isometric to Fα .
Proof. Since α is a dilation norm of E, by Proposition 2.27, Sα and Tα both
are invertible operators with Sα Tα = IX and Tα Sα = IM
fE,α . For any x ∈ X, we
have
kTα xkα = kEΩ,x kα = sup kE(B)xk ≤ sup kE(B)k kxk = kxk,
B∈Σ
B∈Σ
and thus kTα k ≤ 1.
N
N
On the other hand, for any N > 0, {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and {Bi }i=1 ⊂ Σ,
we have that
N
!
N
X
X
Ci E(Bi )xi Ci EBi ,xi = Sα
i=1
i=1
X
X
N
N
X
X
0
Ci EBi ,xi ,
Ci E (Bi ∩ B ) xi = ≤ sup B 0 ∈Σ i=1
i=1
X
α
and so kSα k ≤ 1.
Since Sα Tα = IX and Tα Sα = IM
fE,α , we have that for each x ∈ X we have
||x|| = ||Sα Tα x|| ≤ ||Tα x|| ≤ ||x||.
Hence Tα is a surjective isometry. Similarly Sα is also a surjective isometry, and
therefore E is isometric to Fα .
Proposition 2.34. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system and α be the minimal dilation norm of E. If Pj : Y → Y are norm one
PM
projections such that j=1 Pj = IY , then
Ej (B) : Σ → B(X, Y ),
x∈X
PN
is an operator-valued measure for 1 ≤ j ≤ M . Moreover for any i=1 Ci EBi ,xi ∈
ME , we have
N
N
N
X
X
X X
max Ci (Ej )Bi ,xi ≤ Ci EBi ,xi ≤
Ci (Ej )Bi ,xi ,
1≤j≤M i=1
αj
i=1
Ej (B)x = Pj E(B)x,
α
1≤j≤M
i=1
αj
where αj is the minimal dilation norm of Ej .
Proof. It is obvious that for each 1 ≤ j ≤ M,
Ej (B) : Σ → B(X, Y ),
is an operator-valued measure.
Ej (B)x = Pj E(B)x,
x∈X
2.4. THE MINIMAL DILATION NORM k · kα
33
The “moreover” part follows from:
N
N
X
X
0
= sup Ci Pj E (Bi ∩ B ) xi Ci (Ej )Bi ,xi 0
B ∈Σ i=1
i=1
αj
!
N
X
0
= sup Pj
Ci E (Bi ∩ B ) xi 0
B ∈Σ
i=1
N
X
≤ kPj k sup Ci E (Bi ∩ B 0 ) xi 0
B ∈Σ i=1
N
X
= sup Ci E (Bi ∩ B 0 ) xi B 0 ∈Σ i=1
!
X
N
X
M
0
Pj
Ci E (Bi ∩ B ) xi = sup B 0 ∈Σ j=1
i=1
M
N
X
X
≤
sup Ci Pj (E (Bi ∩ B 0 ) xi )
0 ∈Σ B
j=1
i=1
N
X X
=
Ci (Ej )Bi ,xi .
1≤j≤M
i=1
αj
Lemma 2.35. If {i} is an atom in Σ, then the rank of Fα ({i}) is equal to the
rank of E({i}).
Proof. Suppose that the rank of E({i}) is k. Then there exist x1 , ..., xk ∈ X
such that range E({i}) = span{E({i})x1 , ..., E({i})xk }. We show that range F ({i}) =
span{Ex1 ,{i} , ..., Exk ,{i} }. In fact, for any x ∈ X, B ∈ Σ, we have that F ({i})EB,x =
E{i},x or 0. Now for any A ∈ Σ, we have E{i},x (A) = E({i})x if i ∈ A and 0 if i ∈
/ A.
Pk
Pk
Write E({i})x = j=1 cj E({i})xj . Then we get E{i},x (A) = j=1 cj E({i})xj =
Pk
j=1 cj E{i},xj (A) if i ∈ A and 0 otherwise. Hence
E{i},x =
k
X
cj E{i},xj ,
j=1
and therefore range F ({i}) = span{E{i},x1 , ..., E{i},xk , } as claimed.
Problem A. Is it always true that with an appropriate notion of rank function
for an operator valued measure, that r(F (B)) = r(E(B)) for every B ∈ Σ? The
previous lemma tells us that it is true when B is an atom. A possible “rank ”
definition might be: r(B) = sup{rankE(A) : A ⊂ B, A ∈ Σ}.
Example 2.36. Assume that (N, 2N , E, B(X)) is an induced operator-valued
P
probability measure system by a framing {xi , yi }i∈N of X, where E(B) = i∈B xi ⊗
yi for every B ∈ 2N .
34
2. DILATION OF OPERATOR-VALUED MEASURES
fE,α . Let (N, 2N , Fα , B(M
fE,α ))
We characterize its minimal α-dilation space M
be the corresponding spectral operator-valued probability measure. Then we have
f=
X
fE,α .
Fα ({i})f, ∀ f ∈ M
i∈N
By Lemma 2.35, all Fα ({i})’s are rank-one projections. Choose x̃i such that
yi (x̃i ) = 1. Then
Fα ({i}) E{i},x̃i = E{i},x̃i ,
E{i},x̃ = kxi k, ∀ i ∈ N.
i α
And because Fα ({i})’s are projections with Fα ({i})Fα ({j}) = δij Fα ({i}), we know
fE,α , which is just our minimal framing model basis.
that {E{i},x̃i } is a basis of M
Actually, for every (ai ) ∈ c00 , we have
X
ai E{i},x̃i i∈N
=
α
=
=
=
=
X
ai E({i} ∩ B)x̃i sup B⊂N i∈N
X
sup ai E({i})x̃i B⊂N i∈B
X
sup ai (xi ⊗ yi )x̃i B⊂N
i∈B
X
sup ai yi (x̃i )xi B⊂N
i∈B
X
sup ai xi ,
B⊂N i∈B
which is equivalent to the minimal framing model basis because of the following
inequality.
X
X
X
1
i ai xi .
sup i ai xi ≤ sup ai xi ≤ sup i =±1 B⊂N 2 i =±1 i∈N
i∈B
i∈N
2.4. THE MINIMAL DILATION NORM k · kα
This is because
X
1
i ai xi =
sup 2 i =±1 i∈N
≤
=
=
≤
=
35
X
1
X
ai xi −
ai xi sup 2 i =±1 i∈N
i∈N
=1
i =−1
i
!
X
X
1
ai xi ai xi + sup sup sup B⊂N 2 i =±1 B⊂N i∈B
i∈B
X
ai xi sup B⊂N i∈B
X
1
X
i a i x i +
i ai xi sup 2 B⊂N i∈N
i =1,i∈B
=1
i =−1,i6∈B
i
X
i ai xi sup sup B⊂N i =±1
i∈N
X
sup i ai xi .
i =±1 i∈N
The above example shows that in the case that the operator-valued probability measure system is induced by a framing {xi , yi }i∈N of X, then the minimal
α-dilation space is exactly the one constructed in [CHL] and hence it is always
separable. However, it is not clear whether this is true in general. So we ask:
Problem B. Assume that (Ω, Σ, E, B(X)) is an operator-valued measure system
such that X is a separable Banach space. Is the dilation space α equipped with the
minimal dilation norm always separable? If not, does there exist a dilation space
which is separable?
We end this section with a result concerning this problem. We say that an
operator valued measure E on (Ω, Σ) into B(X) is totally strongly countably additive
if whenever {Bn } is a disjoint sequence in Σ with union B, and x ∈ X, then given
any > 0, there exists N ∈ N such that
∞
X
supA∈Σ ||
E(Bj ∩ A)x|| < j=n
holds for all n ≥ N . We say that (Ω, Σ, E) has a countable Borel basis ifPthere exists
∞
a countable set B of subsets from Σ such that for each B ∈ Σ, B = n=1 Bn for
some subsets Bn from B up to a E-null set (A E-null set is a set C ∈ Σ) such that
E(A) = 0 if A ⊆ C and A ∈ Σ).
Proposition 2.37. Assume that an operator valued measure E on (Ω, Σ) into
B(X) is totally strongly countably additive, and that (Ω, Σ, E) has a countable
Borel basis. If X is separable, then so is the dilation space α with the minimal
dilation norm.
Proof. Fix x ∈ X, and let {Bi } be a countable Borel basis for (Ω, Σ, E).
Then it is easy to prove that span{EBi ,x : i ∈ N} contains EB,x for every B ∈ Σ.
36
2. DILATION OF OPERATOR-VALUED MEASURES
Now fix B ∈ Σ, and let {xi } be a dense subset of X. Then span{EB,xi : i ∈ N}
contains EB,x for every x ∈ X. Therefore we get that span{EBi ,xj : i, j ∈ N} is
dense in α and hence α is separable.
2.5. The Dual Space of the Minimal Dilation Space
In this section we give a concrete description for the dual space of the α-dilation
space. Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system, Y ∗ be the
f∗ be the dual space of M
fE,α . For any y ∗ ∈ Y ∗ and B ∈ Σ,
dual space of Y and M
E,α
fE,α → C by
define a mapping φB,y∗ : M
!
!
N
N
X
X
∗
φB,y∗
Ci EBi ,xi = y
Ci E(B ∩ Bi )(xi ) .
i=1
If
PN
Ci EBi ,xi = 0, then
!
N
X
Ci EBi ,xi = y ∗
i=1
i=1
φB,y∗
i=1
N
X
!
Ci E(B ∩ Bi )(xi )
=y
i=1
∗
N
X
!
Ci EBi ,xi (B)
= 0.
i=1
fE,α → C is well-defined.
Thus the mapping φB,y∗ : M
Proposition 2.38. Let ME = span{φB,y∗ : y ∗ ∈ Y ∗ , B ∈ Σ}. Then ME ⊂
f∗ .
M
E,α
Proof. It is sufficient to prove that φB,y∗ is bounded on ME,α . In fact, this
follows immediately from the following:
!
N
X
Ci EBi ,xi φB,y∗
i=1
!
N
∗ X
= y
Ci E(B ∩ Bi )(xi ) i=1
N
X
≤ ky ∗ k Ci E(Bi ∩ B)xi i=1
N
X
∗ ≤ ky k Ci EBi ,xi .
i=1
α
f∗ by α∗ . Then the space ME endowed with
We will denote the norm on M
E,α
∗
the norm α is a normed space, which will be denoted by ME,α∗ .
PN
Proposition 2.39. For every i=1 Ci φBi ,yi∗ ∈ ME,α∗ , we have
N
N
X
X
Ci φBi ,yi∗ ≤
|Ci |kyi∗ k.
∗
i=1
α
i=1
Proof. First, for every φy∗ ,B ∈ ME,α∗ , by the proof of Proposition 2.38, we
have
kφB,y∗ k ≤ ky ∗ k.
2.6. THE MAXIMAL DILATION NORM k · kω
37
PN
Thus, for every i=1 Ci φBi ,yi∗ ∈ ME,α∗ , we get
N
N
N
X
X
X
Ci φBi ,yi∗ ≤
|Ci | φBi ,yi∗ ≤
|Ci |kyi∗ k.
∗
i=1
i=1
α
i=1
Since α∗ is a norm on ME , we denote the completion of ME under norm α∗
fE,α∗ . Then M
fE,α∗ is a Banach space, and so M
fE,α∗ ⊂ M
f∗ .
by M
E,α
fE,α∗
f∗ = M
Proposition 2.40. M
E,α
w∗
.
Proof. By the Hahn-Banach Separation Theorem with respect to the w∗ fE,α∗ separates M
fE,α . If not, then there is
topology, it is enough to show that M
∗
f
f∗ . For any
an f ∈ ME,α with kf kα = 1 such that g (f ) = 0 for every g ∗ ∈ M
E,α
0 < < 1/2, we can find a g ∈ ME,α with kgkα = 1 and kf − gkα ≤ . Let
PN
g = i=1 Ci EBi ,xi be a representation of g. By the definition of α-dilation norm,
there is E ∈ Σ satisfying
N
X
Ci E(Bi ∩ B)(xi ) ≥ 1 − .
i=1
Take an y ∗ ∈ Y ∗ such that
y∗
N
X
!
Ci E(Bi ∩ B)xi
N
X
Ci E(Bi ∩ B)xi ≥ 1 − .
=
i=1
i=1
Then for φB,y∗ ∈ ME,α∗ ,
φB,y∗ (g) = φB,y∗
N
X
!
Ci EBi ,xi
=y
∗
i=1
N
X
!
Ci E(Bi ∩ B)xi
≥1−
i=1
and kφB,y∗ kα∗ ≤ ky ∗ k = 1. Thus, we have
φB,y∗ (f )
= φB,y∗ (f − g) + φB,y∗ (g)
≥ 1 − − kφB,y∗ kα∗ kf − gk
≥ 1 − 2
> 0,
fE,α∗
f∗ = M
which leads to a contradiction. Thus M
E,α
w∗
.
2.6. The Maximal Dilation Norm k · kω
In this section, we construct the “maximal ” dilation norm for the elementary
dilation space ME . Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system.
Consider the basic elements EB,x ∈ ME . It is natural to require that
(2.5)
kEB,x k ≤ sup kE(B ∩ B 0 )xk .
B 0 ∈Σ
38
2. DILATION OF OPERATOR-VALUED MEASURES
PN
Now let f be any element of ME . If i=1 Ci EBi ,xi is a representation of f , then
it follows from the triangle inequality that
kf k ≤
N
X
sup kCi E(Bi ∩ B)xi k.
i=1 B∈Σ
Since this holds for every representation of f , we get
N
nX
o
kf k ≤ inf
sup kCi E(Bi ∩ B)xi k ,
i=1 B∈Σ
where the infimum is taken over all representation of f . Define k · kω : ME →
R+ ∪ {0} by
(N
)
N
X
X
kf kω = inf
sup kCi E(Bi ∩ B)xi k : f =
Ci EBi ,xi ∈ ME .
i=1 B∈Σ
i=1
Then we have the following proposition.
Proposition 2.41. k · kω is a semi-norm on ME and
kEB,x kω = sup kE(B ∩ B 0 )xk
E 0 ∈Σ
for every B ∈ Σ and x ∈ X.
Proof. First, we show that kλf kω = |λ| kf kω . This is obvious when λ is zero.
PN
So suppose that λ 6= 0. If f = i=1 Ci EBi ,xi is a representation of f , then
λf =
N
X
λCi EBi ,xi =
N
X
Ci EBi ,λxi ,
i=1
i=1
and so we have
kλf kω ≤
N
X
sup kCi E(Bi ∩ B)(λxi )k = |λ|
i=1 B∈Σ
N
X
sup kCi E(Bi ∩ B)xi k.
i=1 B∈Σ
Since this holds for every representation
of f , it follows that kλf kω ≤ |λ| kf k. In
the same way, we have kf kω = λ−1 λf ω ≤ |λ|−1 kλf kω , giving |λ| kf kω ≤ kλf kω .
Therefore kλf kω = |λ| kf kω .
Now, to prove that k · kω satisfies the triangle inequality. Let f, g ∈ ME
and let > 0. It follows from the definition that we may choose representations
PN
PM e
f = i=1 Ci EBi ,xi and g = j=1 C
j Eyj ,Aj such that
N
X
sup kCi E(Bi ∩ B)xi k ≤ kf kω + /2,
i=1 B∈Σ
M
X
Then
ej E(Aj ∩ A)xj k ≤ kgkω + /2.
sup kC
j=1 A∈Σ
PN
PM
i=1 Ci EBi ,xi +
j=1
kf +gkω ≤
N
X
ej EA ,y is a representation of f + g and so
C
i i
sup kCi E(Bi ∩B)xi k+
i=1 B∈Σ
M
X
ej E(Aj ∩A)xj k ≤ kf kω +kgkω +.
sup kC
j=1 A∈Σ
Since this holds for every > 0, we have kf + gkω ≤ kf kω + kgkω .
2.6. THE MAXIMAL DILATION NORM k · kω
39
Finally, we must show that kEB,x kω = supB 0 ∈Σ kE(B ∩ B 0 )xk. On the one
hand, it is clear that kEB,x kω ≤ supB 0 ∈Σ kE(B ∩ B 0 )xk. On the other hand, if
PN
EB,x = i=1 Ci EBi ,xi is a representation of EB,x , we have
sup kE(B ∩ B 0 )xk
N
X
sup Ci E(Bi ∩ B 0 )xi =
B 0 ∈Σ
B 0 ∈Σ
≤
sup
i=1
N
X
B 0 ∈Σ i=1
≤
N
X
kCi E(Bi ∩ B 0 )xi k
sup kCi E(Bi ∩ B 0 )xi k.
0
i=1 B ∈Σ
Since this holds for every representation of EB,x , it follows that supB 0 ∈Σ kE(B ∩
B 0 )xk ≤ kEB,x kω . Therefore kEB,x kω = supB 0 ∈Σ kE(B ∩ B 0 )xk.
We define an equivalence relation Rω on ME by f ∼ g if kf − gkω = 0. Then
k · kω is a norm on ME . Denote by ME,ω the space of the Rω -equivalence classes of
fE,ω for its completion.
ME endowed with the norm k · kω , and by M
Theorem 2.42. k · kω is a dilation norm of E.
N
N
Proof. Let N > 0, {Ci }N
i=1 ⊂ C, {xi }i=1 ⊂ X and {Bi }i=1 ⊂ Σ. First, we
fE,ω → Y defined on ME,ω by
show that the map Sω : M
!
N
N
X
X
Sω
Ci EBi ,xi =
E(Bi )xi ,
i=1
i=1
PN
is well-defined and kSω k ≤ 1. If f = i=1 Ci EBi ,xi is a representation of f, then
N
N
X
X
Ci E(Bi )xi ≤
sup kCi E(Bi ∩ B)xi k.
kSω (f )k = B∈Σ
i=1
i=1
Since this holds for every representation of f , it follows that kSω (f )k ≤ kf kω .
Therefore Sω is well-defined and bounded with kSω k ≤ 1.
fE,ω with T (x) = EΩ,x is bounded
Now, to prove that the map Tω : X → M
with kTω k ≤ kEk. It follows from the definition of ω that
kTω xkω = kEΩ,x kω = sup kE(B)xk ≤ kEk · kxk.
B∈Σ
Thus, kTω k ≤ kEk.
fE,ω ) defined by
Finally we need to prove that the map Fω : Σ → B(M
!
N
N
X
X
Fω (B)
Ci EBi ,xi =
Ci EB∩Bi ,xi
i=1
i=1
is an operator-valued measure. By Lemma 2.25, we only need to show that Fω
is strongly countably additive and uniform bounded on ME,ω . The proof is very
similar to that in Theorem 2.29. We include it here for completeness.
40
2. DILATION OF OPERATOR-VALUED MEASURES
If f =
PN
Ci EBi ,xi is a representation of f , then
!
N
X
Ci EBi ,xi kFω (B) (f )kω = Fω (B)
i=1
ω
N
X
= Ci EBi ∩B,xi i=1
i=1
≤
N
X
ω
sup kCi E(Bi ∩ B ∩ B 0 )xi k
0
i=1 B ∈Σ
=
N
X
sup kCi E(Bi ∩ B 0 )xi k.
0
i=1 B ∈Σ
Since this holds for every representation of f , it follows that kFω (E)f kω ≤
kf kω , which implies that kFω (B)k ≤ 1.
For the strong countably additivity of Fω , let {Aj }∞
j=1 be a disjoint countable
collection of members of Σ with union A. Then
!
!
M
N
N
X
X
X
Fω (Aj )
Ci EBi ,xi − Fω (A)
Ci EBi ,xi j=1
i=1
i=1
ω
!
!
M
N
N
X
X X
Ci EAj ∩Bi ,xi −
Ci EA∩Bi ,xi = j=1 i=1
i=1
ω


N
M
X
X
Ci 
EAj ∩Bi ,xi − EA∩Bi ,xi 
= i=1
j=1
ω


N
M
X
X
0
0


C
E(A
∩
B
∩
B
)x
−
E(A
∩
B
∩
B
)x
= sup i
j
i
i
i
i B 0 ∈Σ i=1
j=1
Y

 N
∞
[
X
= sup Ci E 
(Aj ∩ Bi ∩ B 0 ) xi B 0 ∈Σ i=1
j=M +1

 Y
N
∞
X
[
0 

≤
|Ci | sup E
(A
∩
B
∩
B
)
x
j
i
i
B 0 ∈Σ i=1
j=M +1
Y

 N
∞
X
[
=
|Ci | sup E
(Aj ∩ B 0 ) xi B 0 ∈Σ i=1
j=M +1
Y
∞
N
X
X
=
|Ci | sup E(Aj ∩ B 0 )xi .
0
B ∈Σ i=1
j=M +1
Y
P
∞
If supB 0 ∈Σ j=M +1 E(Aj ∩ B 0 )xi does not tend to 0 when M tends to ∞,
Y
then we can find δ > 0, a sequence of n1 ≤ m1 < n2 ≤ m2 < n3 ≤ m3 < . . . , and
2.6. THE MAXIMAL DILATION NORM k · kω
41
{Bl0 }∞
l=1 ⊂ Σ such that
X
ml
0
E(A
∩
B
)x
j
l i ≥ δ, ∀ l ∈ N.
j=nl
Obviously, for l ∈ N and nl ≤ j ≤ ml , Aj ∩ Bl0 are disjoint from each other, so


ml
ml
∞ [
∞ X
[
X
E
Aj ∩ Bl0  xi =
E(Aj ∩ Bl0 )xi ,
l=1 j=nl
l=1 j=nl
P
ml
0
E(A
∩
B
)x
which implies j=n
→ 0. It is a contradiction, hence we have
j
i
l
l
!
!
N
∞
N
X
X
X
Ci EBi ,xi .
Fω (Aj )
Ci EBi ,xi =
Fω (A)
j=1
i=1
i=1
Thus Fω is strongly countably additive.
In what follows we will refer ω as the maximal dilation norm of E. From
Proposition 2.41 we have that both the minimal and the maximal dilation norms
agree on the elementary vectors Ex,B , i.e., ||Ex,B ||α = ||Ex,B ||ω .
Finally we point out that for some special operator-valued measure systems,
there are some other natural ways to construct new dilations norms. We mention
two of them for which we will only give the definition but will skip the proofs.
Let (Ω, Σ, E, B(X, Y )) be an operator-valued measure system. The strong variation of E is the extended nonnegative function |E|SOT whose value on a set B ∈ Σ
is given by
nX
o
|E|SOT (B) = sup
kE(Bi )xk : x ∈ BX , Bi ’s are a partition of B
=
sup |Ex |(B).
x∈X
If |E|SOT (Ω) < ∞, then E is called an operator-valued measure of strongly bounded
variation. Similarly, the weak variation of E is the extended nonnegative function
|E|WOT whose value on a set B ∈ Σ is given by
nX
o
|E|WOT (B) = sup
|x∗ (E(Bi )x)| : x ∈ X, y ∗ ∈ Y ∗ , Bi ’s are a partition of B
=
sup
x∈X,y ∗ ∈Y ∗
|Ex,y∗ |(B).
If |E|WOT (Ω) < ∞, then E is called an operator-valued measure of weakly bounded
variation.
For an operator-valued measure of strongly bounded variation (respectively, of
weakly bounded variation) , we have a natural approach to construct a dilation
PN
e (f ) and
norm on ME . Now let f = i=1 Ci EBi ,xi be any element of ME . Define ω
f
W(f ) by
ω
e (f )
=
M X
N
nX
o
sup
Ci E(Bi ∩ Aj )xi : Aj ’s are a partition of Ω
j=1
= |Ef |(Ω).
i=1
42
2. DILATION OF OPERATOR-VALUED MEASURES
and
f )
W(f
=

M X
∗
sup
y

j=1
=
N
X
i=1

!

Ci E(Bi ∩ Aj )xi : y ∗ ∈ Y ∗ , Aj ’s are a partition of Ω

∗
sup |y Ef |(Ω)
y ∗ ∈Y ∗
= kEf k(Ω).
f is a dilation norm of E.
Then it can be proved that ω
e (resp. W)
CHAPTER 3
Framings and Dilations
We examine the dilation theory for discrete and continuous framing-induced
operator valued measures. We also provide a new self-contained proof of Naimark’s
Dilation Theorem based on our methods of chapter 2, because we feel that this
helps clarify our approach, and also for independent interest.
3.1. Hilbertian Dilations
In the Hilbert space theory, the term “projection” is usually reserved for “orthogonal” (i.e. self-adjoint) projection. So in this Chapter we will resume this
tradition. The term “idempotent” will be used to denote a not necessarily selfadjoint operator which is equal to its square. Throughout this chapter, H denotes
the Hilbert space and IH is the identity operator on H. Let 2N denote the family
of all subsets of N.
Definition 3.1. Given an operator-valued measure E : Σ → B(H). Then E
is called:
(i) an operator-valued probability measure if E(Ω) = IH ,
(ii) a (orthogonal) projection-valued measure if E(B) is a (orthogonal) projection on H for all B ∈ Σ; an idempotent-valued measure if E is a spectral
in the sense of Definition 2.5,
(iv) a positive operator-valued measure if E(B) is a positive operator on H
for all B ∈ Σ.
Let (Ω, Σ, E, B(H)) be an operator-valued measure system. The definition of
dilation space is the same as in section 2.2. Keep in mind that not every operatorvalued measure on a Hilbert space admits a dilation space which is a Hilbert space.
Definition 3.2. Let (Ω, Σ, E, B(H)) be an operator-valued measure system.
We say that E has a Hilbertian dilation if it has a dilation space which is a Hilbert
space. That is, E has a Hilbert dilation space if there exist a Hilbert space K,
two bounded linear operators S : K → H and T : H → K, an idempotent-valued
measure F : Σ → B(K) such that
E(B) = SF (B)T,
∀B ∈ Σ.
Lemma 3.3. Let (Ω, Σ, E, B(H)) be an operator-valued probability measure system. If E has a Hilbert dilation space, then there exist a corresponding Hilbert dilation system (Ω, Σ, F, B(K), V ∗ , V ) such that V : H → K is an isometric embedding.
Proof. From Proposition 2.16, we know that if an operator-valued measure
can be dilated to a spectral operator-valued measure, then it can be dilated to a
spectral operator-valued probability measure. So without losing the generality, we
can assume that the corresponding dilation projection-valued measure space system
43
44
3. FRAMINGS AND DILATIONS
(Ω, Σ, F, B(K), S, T ) is a probability measure space system, where K is the Hilbert
dilation space.
Since E and F are both operator-valued probability measures, we have that S
is a surjection, T is an isomorphic embedding and IH = ST. Let P = T S : K → T H.
Then P is a projection from K onto T H. Hence
K = P K ⊕K (IK − P )K = T H ⊕K (IK − P )K
From
S(IK − P ) = S − ST S = S − S = 0,
we get that
(IK − P )K ⊂ kerS.
On the other hand, for z ∈ K, if Sz = 0, then
(IK − P )z = z − P z = z − T Sz = z ∈ (IK − P )K
Thus
K = T H ⊕K kerS.
Define
e = H ⊕2 kerS.
H
It is easy to see that the operator
e = H ⊕2 kerS
U : K = T H ⊕K kerS → H
defined by
U = T −1 |T H ⊕ IkerS ,
e For any
is an isomorphic operator. Let V = U T be an operator form H → H.
x ∈ H, we have
kV xkHe = kU T xkHe = kxkH .
Thus V is an isometric embedding.
Define
e
F̃ : Σ → B(H),
F̃ (B) = U F (B)U −1 .
Then F̃ is a spectral operator-valued probability measure. Now we show that
E(B) = V ∗ F̃ (B)V,
∀B ∈ Σ.
Since
V ∗ F̃ (B)V = V ∗ U F (E)U −1 V = (U T )∗ U F (B)U −1 U T = (U T )∗ U F (B)T,
we only need to prove that (U T )∗ U = S, as claimed.
For any x ∈ H and z = T x1 + z2 ∈ K, where z2 ∈ kerS and x1 ∈ H,
h(U T )∗ U z, xiH = hU z, U T xiHe = hU T x1 + U z2 , xiHe = hx1 , xiH
and
hSz, xiH = hST x1 + Sz2 , xiH = hx1 , xiH
Hence (U T )∗ U = S.
3.1. HILBERTIAN DILATIONS
45
The classical Naimark’s Dilation Theorem tells us that if E is a positive operatorvalued measure, then E has a Hilbert dilation space. Moreover, the corresponding
spectral operator-valued measure F is an orthogonal projection-valued measure.
Although this is a well known result in the dilation theory, for self completeness
here we include a new proof which uses similar line of ideas as used in the proof of
the Banach space dilation theory.
Theorem 3.4 (Naimark’s Dilation Theorem). Let E : Σ → B(H) be a
positive operator-valued measure. Then there exist a Hilbert space K, a bounded
linear operator V : H → K, and an orthogonal projection-valued measure F : Σ →
B(K) such that
E(B) = V ∗ F (B)V.
Proof. Let ME = span{EB,x : x ∈ H, B ∈ Σ} be the space induced by
(Ω, Σ, E, B(H)). Now we define a sesquilinear functional h , i on this space by setting
+
*N
N X
M
M
X
X
X
hE(Bi ∩ Aj )xi , yj iH .
EAi ,yi =
EBi ,xi ,
i=1 j=1
j=1
i=1
PN
PN
Actually, if i=1 EBi ,xi = 0, then for 1 ≤ j ≤ M , i=1 E(Bi ∩ Aj )xi = 0. Thus
PN
this sesquilinear functional is well-defined. Since for any f = i=1 EBi ,xi (without
losing the generality, we can assume that Bi ’s are disjoint from each other), we
have that
*N
+
N
N
N
N X
X
X
X
X
hE(Bi ∩ Bj )xi , xj iH =
hE(Bi )xi , xi iH ≥ 0.
EBi ,xi ,
EBj ,xj =
i=1
i=1 j=1
j=1
i=1
Thus it follows that h , i is positive definite.
Obviously, this positive definite sesquilinear functional satisfies the CauchySchwarz inequality,
|hf, gi|2 ≤ hf, f i · hg, gi.
fE denote
Hence the sesquilinear functional h , i on ME is an inner product. Let M
the Hilbert space that is the completion of the inner product space ME , and the
induced norm is denoted by k · kM
fE .
fE → M
fE by
For every B ∈ Σ, define a linear map F (B) : M
!
N
N
X
X
F (B)
EBi ,xi =
Exi ,B∩Bi .
i=1
i=1
PN
fE , then
It is easy to see that F (B) is a projection. Take i=1 EBi ,xi ∈ M
!
N
N
X
X
EBi ,xi =
hE(B ∩ Bi )xi , xi iH
F (B)
i=1
i=1
fE
M
≤
N
X
hE(Bi )xi , xi iH
i=1
N
X
= EBi ,xi i=1
fE
M
.
46
3. FRAMINGS AND DILATIONS
So kF (B)k = 1 or kF (B)k = 0. Moreover, we have
! M
*
+
N
N X
M
X
X
X
F (B)
EBi ,xi ,
EAj ,yj
=
hE(B ∩ Bi ∩ Aj )xi , yj iH
i=1
j=1
i=1 j=1
fE
M
=
*N
X

+
M
X
EBi ,xi , F (B) 
EAj ,yj 
i=1
j=1
.
fE
M
Thus F (E) is a self-adjoint orthogonal projection.
Now we define
fE ,
V :H→M
V (x) = Ex,Ω .
Since
2
kV xk2M
f = hE(Ω)x, xiH ≤ kE(Ω)k · kxkH ,
E
we know that V is bounded. Indeed, it is clear that
kV k2 = sup{hE(Ω)x, xiH : kxkH ≤ 1} = kE(Ω)k.
Observe that for any x, y ∈ H,
hV ∗ F (B)V x, yiH = hF (B)(EΩ,x ), V (EΩ,y )iM
fE = hE(B)x, yiH ,
so E(B) = V ∗ F (B)V , which completes the proof.
For general Hilbert space operator-valued measures, Don Hadwin completely
characterized those that have Hilbertian dilations in terms of Hahn decompositions.
Let Ω be a compact Hausdorff space and Σ be the σ-algebra of Borel subsets of
Ω. Recall that a B(H)-valued measure E is called regular if for all x, y ∈ H the
complex measure given by < E(·)x, y > is regular.
Theorem 3.5. ([HA]) Let E be a regular, bounded B(H)-valued measure.
Then the following are equivalent:
(i) E has a Hahn decomposition E = (E1 − E2 ) + i(E3 − E4 ), where Ej (j =
1, 2, 3, 4) are positive operator-valued measures;
(ii) there exist a Hilbert space K, two bounded linear operators S : K → H and
T : H → K, an projection-valued measure F : Σ → B(K) such that
E(B) = SF (B)T,
∀B ∈ Σ.
3.2. Operator-Valued Measures Induced by Discrete Framings
Let {xi }i∈N be a non-zero frame for a separable Hilbert space H. Then the
mapping E : N → B(H) defined by
X
E(B) =
xi ⊗ xi , ∀ B ∈ 2 N ,
i∈B
is a positive operator-valued probability measure. By Theorem 3.4, we know
fE , the corresponding operator-valued meathat E have a Hilbert dilation space M
sure F is self-adjoint and idempotent. Obviously, the rank of F ({i}) is 1, so
fE . For any
{E{i},xi /kxi k2 }i∈N is an orthonormal basis of the Hilbert space M
N
B ∈ 2 , if i ∈ B, then
E{i},x (B) = E(B ∩ {i})x = E({i})x = hx, xi ixi ,
3.2. OPERATOR-VALUED MEASURES INDUCED BY DISCRETE FRAMINGS
hx, xi i
47
E{i},xi (B)
hx, xi i
hx, xi i
=
E(B ∩ {i})xi =
hxi , xi ixi = hx, xi ixi .
kxi k2
kxi k2
kxi k2
E
{i},xi
∗
is
Thus E{i},x = hx, xi i kx
2 . Then V is the traditional analysis operator and V
ik
the traditional synthesis operator. This is because
X
E{i},xi
V x = Ex,N =
hx, xi i
kxi k2
i∈N
and
V
∗
X
i∈N
E{i},xi
ai
kxi k2
!
=
X
ai
i∈N
X
E({i})xi
=
ai xi .
kxi k2
i∈N
Definition 3.6. Let {xi }i∈N and {yi }i∈N are both frames for H. We call
{xi }i∈N and {yi }i∈N are scale equivalent if for each i ∈ N, yi = λi xi for some
λi ∈ C. {xi }i∈N and {yi }i∈N are called unit-scale equivalent if we can take |λj | = 1
for all i ∈ N.
Lemma 3.7. Two frames {xi }i∈N and {yi }i∈N are unit-scale equivalent if and
only if {xi }i∈N and {yi }i∈N generate the same positive operator-valued measure.
The following theorem shows exactly when a framing induced operator-valued
measure has a Hilbertian dilation.
Theorem 3.8. Let (xi , yi )i∈N be a non-zero framing for a Hilbert space H. Let
E be the induced operator-valued probability measure, i.e.,
X
E(B) =
xi ⊗ yi ,
∀ B ⊆ N.
i∈B
Then E has a Hilbert dilation space K if and only if there exist αi , βi ∈ C, i ∈ N
with αi β̄i = 1 such that {αi xi }i∈N and {βi yi }i∈N both are the frames for the Hilbert
space H.
Proof. We first prove the sufficient condition. If E has a Hilbert dilation space
K, then by Theorem 2.26, there is an elementary dilation operator-valued measure
fE,D ), SD , TD ) of E. Since the norm on ME,D is defined by
system (Ω, Σ, FD , B(M
X
X
EBi ,xi = F (Bi )T (xi ) ,
i
D
i
K
fE,D is a Hilbert space.
M
Recall that the analysis operator
fE,D ,
TD : H → M
TD (x) = Ex,Ω
and the synthesis operator
!
fE,D → H,
SD : M
SD
X
EBi ,xi
i
=
X
E(Bi )xi
i
are both linear and bounded. It is easy to prove that FD ({i}) is rank 1 for all i ∈ N.
fE,D such that
By Lemma 2.12, there is a Riesz basis φi on M
FD ({i}) = φi ⊗ φ∗i .
48
3. FRAMINGS AND DILATIONS
Hence
xi ⊗ yi = SD (φi ⊗ φ∗i ) TD = (SD φi ) ⊗ (TD∗ φ∗i ).
For any x ∈ H, we have
X
X
2
2
∗
|hx, SD φi i| =
|hSD
x, zi i| ,
i∈N
i∈N
X
2
|hx, TD∗ φ∗i i| =
i∈N
Thus we know that {SD φi }i∈N and
{TD∗ φ∗i }i∈N
X
2
|hTD x, φ∗i i| .
i∈N
both are frames for H and
xi ⊗ yi = (SD φi ) ⊗ (TD∗ φ∗i ).
So there is only a scalar adjustment SD φi = αi xi and TD∗ φ∗i = βi yi with αi β̄i = 1.
For the necessary part, assume that there exist αi , βi ∈ C, i ∈ N with αi β̄i = 1
such that {αi xi }i∈N and {βi yi }i∈N both are the frames for the Hilbert space H. By
Proposition 1.6 in [HL], we have a Hilbert space K, a Riesz basis {zi }i∈N and a
projection P : K → H such that P zi = αi xi . Let {zi∗ }i∈N be the dual Riesz basis,
then hzi , zj∗ i = δi,j . Define S = P
T :H→K
T ∗ zi∗ = βi yi
and
F (B) = zi ⊗ zi∗
F : 2N → B(K)
∀B ∈ 2N .
Then T is a linear and bounded operator and F is a spectral operator-valued
measure. For any x ∈ H and B ∈ 2N ,
X
X
SF (E)T x = S
hT x, zi∗ izi =
hx, T ∗ zi∗ iP zi
i∈B
=
X
hx, βi yi iαi xi =
i∈B
i∈B
X
hx, yi ixi
i∈B
= E(B)
Thus K is the Hilbert dilation space of E.
Since a completely bounded map is a linear combination of positive maps, the
following Corollary immediately from Theorem 3.5 and Theorem 3.8.
Corollary 3.9. Let {xi , yi }i∈N be a non-zero framing for a Hilbert space H,
and E be the induced operator-valued probability measure. Then E is a completely
bounded map if and only if {xi , yi }i∈N can be re-scaled to dual frames.
By using this theorem, we can prove that the first example of framing in Chapter 5 can not be scaled to a dual frame since the induced operator-valued measure
doesn’t have a Hilbert dilation space,
Lemma 3.10. [Ch, Lemma 5.6.2] Assume that {xi }i∈N and {yi }i∈N are Bessel
sequences in H. Then the following are equivalent:
P
(i) x = Pi∈N hx, yi ixi , ∀x ∈ H.
(ii) x = i∈N
Phx, xi iyi , ∀x ∈ H.
(iii) hx, yi = i∈N hx, xi ihyi , xi i, ∀x, y ∈ H.
In case the equivalent conditions are satisfied, {xi }i∈N and {yi }i∈N are dual frames
for H.
3.2. OPERATOR-VALUED MEASURES INDUCED BY DISCRETE FRAMINGS
49
P Lemma 3.11. [Si, Lemma 14.9] Let {xi }i∈N be a system of vectors in H. If
i∈N xi converges unconditionally in H, then
X
kxi k2 ≤ C < +∞.
i∈N
The following lemma can be proved by using the Orlicz-Pettis Theorem.
Lemma 3.12. A pair {xi , yi }i∈N is a non-zero framing for a Hilbert space H iff
{yi , xi }i∈N is a non-zero framing for H.
The following provides us a sufficient condition under which a framing induced
operator-valued measure has a Hilbertian dilation.
Theorem 3.13. Let {xi , yi }i∈N be a non-zero framing for a Hilbert space H.
If inf kxi k · kyi k > 0, then we can find αi , βi ∈ C, i ∈ N with αi β̄i = 1 such
that {αi xi }i∈N and {βi yi }i∈N both are frames for the Hilbert space H. Hence the
operator-valued measure induced by {xi , yi }i∈N has a Hilbertian dilation.
Proof. Since {xi , yi }i∈N is a non-zero framing for H, we have by Lemma 3.12
that
X
X
x=
hx, yi ixi =
hx, xi iyi ,
i∈N
i∈N
and the series converges unconditionally for all x ∈ H. Applying Lemma 3.11 to
the sequences {hx, yi ixi }i∈N and {hx, xi iyi }i∈N , we get
X
X
2
2
|hx, yi i| kxi k2 < +∞, and
|hx, xi i| kyi k2 < +∞.
i∈N
i∈N
Let
αi = (kyi k/kxi k)1/2
and βi = 1/αi = (kxi k/kyi k)1/2 .
It is easy to see that
X
2
|hx, αi xi i|
=
i∈N
X |hx, xi kyi ki|2
i∈N
≤
kxi kkyi k
X
1
inf kxi k · kyi k
2
|hx, xi i| kyi k2 < +∞
i∈N
and
X
2
|hx, βi yi i|
=
i∈N
X |hx, yi kxi ki|2
i∈N
≤
kxi kkyi k
X
1
inf kxi k · kyi k
2
|hx, yi i| kxi k2 < +∞,
i∈N
Hence {αi xi }i∈N and {βi yi }i∈N are both Bessel sequence. For any x ∈ H, we have
X
X
x=
hx, βi yi iαi xi =
hx, αi xi iβi yi .
i∈N
i∈N
By Lemma 3.11, we conclude that {αi xi }i∈N and {βi yi }i∈N both are frames for
H.
50
3. FRAMINGS AND DILATIONS
Recall from Dai and Larson [DL] that a unitary system U is a subset of the
unitary operators acting on a separable Hilbert space H which contains the identity
operator I. We have the following consequence immediately follows from Theorem
3.13:
Corollary 3.14. Let U1 and U2 be unitary systems on a separable Hilbert
space H. If there exist x, y ∈ H such that {U1 x, U2 y} is a framing of H, then {U1 x}
and {U2 y} both are frames for H.
We remark that there exist sequences {xn } and {yn } in a Hilbert space H with
the following
Pproperties:
(i) x = n hx, xn iyn hold for all x in a dense subset of H, and the convergence
is unconditionally.
(ii) inf ||xn || · ||yn || > 0.
(iii) {xn , yn } is not a framing.
Here is a simple example of this type: Let H = L2 [0, 1], and g(t) = t1/4 ,
f (t) = 1/g(t). Define xn (t) = e2πint f (t) and yn (t) = e2πint g(t). Then it is easy to
verify (i) and (ii). For (iii), we consider the convergence of the series
X
hf, xn iyn .
n∈Z
Note that ||hf, xn iyn ||2 = |hf, xn i|2 · ||g||2 and P
{hf, xn i} is not in `2 (since f 2 ∈
/
2
L [0, 1]). Thus, from Lemma 3.11, we have that n hf, xn iyn can not be convergent
unconditionally. Therefore {xn , yn } is not a framing.
3.3. Operator-Valued Measures Induced by Continuous Frames
We mainly examine the positive-operator valued measure induced by continuous frames investigated by Gabardo and Han in [GH] and also by Fornasier and
Rauhut in [FR]. At the same time, we will introduce a few new “frames” including
operator-valued continuous frames and operator-valued µ-frames, both generalize
the concept of continuous frames.
Definition 3.15. Let H be a separable Hilbert space and Ω be a σ-locally
compact (σ-compact and locally compact) Hausdorff space endowed with a positive
Radon measure µ with suppµ = Ω. A weakly continuous function F : Ω → H is
called a continuous frame if there exist constants 0 < C1 ≤ C2 < ∞ such that
Z
2
C1 kxk ≤
|hx, F(ω)i|2 d µ(ω) ≤ C2 kxk2 , ∀ x ∈ H.
Ω
If C1 = C2 , then the frame is called tight. Associated to F is the frame operator
SF defined by
Z
SF : H → H, hSF (x), yi := hx, F(ω)i · hF(ω), yid µ(ω).
Ω
Then SF is a bounded, positive, and invertible operator. We define the following
transform associated to F,
VF : H → L2 (Ω, µ),
VF (x)(ω) := hx, F(ω)i.
Its adjoint operator is given by
VF∗
2
: L (Ω, µ) → H,
hVF∗ (f ), xi
Z
:=
f (ω)hF(ω), xid µ(ω).
Ω
3.3. OPERATOR-VALUED MEASURES INDUCED BY CONTINUOUS FRAMES
51
A weakly continuous function F : Ω → H is called Bessel if there exist a
positive constant C such that
Z
|hx, F(ω)i|2 d µ(ω) ≤ Ckxk2 , ∀ x ∈ H.
Ω
Let B be the Borel algebra of Ω and F : Ω → H be Bessel. Then the mapping
Z
EF : B → B(H), hEF (B)x, yi =
hx, F(ω)i · hF(ω), yid µ(ω)
B
is well-defined since we have
Z
|hEF (B)x, yi| = hx, F(ω)i · hF(ω), yid µ(ω)
B
Z
≤
|hx, F(ω)i| · |hF(ω), yi|d µ(ω)
Ω
Z
1/2
1/2 Z
2
|hF(ω), yi|2 d µ(ω)
|hx, F(ω)i| d µ(ω)
·
≤
Ω
Ω
≤ C · kxk · kyk.
and therefore kEF (B)k ≤ C for all E ∈ B.
Lemma 3.16. EF defined above is a positive operator-valued measure.
Proof. The positivity follows from that
Z
hEF (B)x, xi =
|hx, F(ω)i|2 d µ(ω) ≥ 0
Ω
holds for all x in H.
Assume that {Bi }i∈N is a sequence of disjoint Borel sets of Ω with union B.
Then we have
X
XZ
hEF (Bi )x, yi =
hx, F(ω)i · hF(ω), yid µ(ω)
i∈N
i∈N
Bi
Z
hx, F(ω)i · hF(ω), yid µ(ω)
=
B
=
hEF (B)x, yi
for all x, y in H. Thus EF weakly countably additive, and hence it is a positive
operator-valued measure.
Lemma 3.17. Let F1 and F2 be two Bessel functions from Ω to H. Then F1
and F2 induce the same positive operator-valued measure if and only if F1 = u · F2
where u is a unimodular function on Ω.
Proof. Since EF1 = EF2 , we have
Z
2
Z
|hx, F1 (ω)i| d µ(ω) = hEF1 (B)x, xi = hEF2 (B)x, xi =
E
|hx, F2 (ω)i|2 d µ(ω)
B
for all B ∈ Σ and x ∈ H. Then by the continuity, we have |hx, F1 (ω)i| = |hx, F2 (ω)i|
for all x ∈ H and ω ∈ Ω. Thus we have
|hF1 (ω), F2 (ω)i| = kF1 (ω)k2 = kF2 (ω)k2 = kF1 (ω)k · kF2 (ω)k.
Then there is a unimodular function on Ω such that F1 (ω) = u(ω) · F2 (ω).
52
3. FRAMINGS AND DILATIONS
fE . The
From Theorem 3.4, we know that EF has a Hilbert dilation space M
F
2
fE is isometric to L (Ω, µ).
following theorem shows that M
F
fE
Theorem 3.18. Suppose that suppF = Ω. Then the dilation Hilbert space M
F
is isometric to L2 (Ω, µ) under the following natural linear surjective isometry
fE → L2 (Ω, µ),
U :M
F
U (EF B,x )(ω) = χB (ω) · hx, F(ω)i
Proof. Without losing the generality, we can assume that {Bi } is a disjoint
countable collection of members of Σ. Then we have
2
X
X
=
hEF (Bi )xi , xi i
EF Bi ,xi f
i
i
MEF
XZ
2
|hxi , F(ω)i| d µ(ω)
=
i
Bi
2
Z X
χBi (ω) · hxi , F(ω)i d µ(ω)
=
Ω i
!2
X
.
EF Bi ,xi = U
2
i
L (Ω,µ)
Thus U is an isometry.
Now we prove that U is surjective. Because all the functions χB (ω) are dense
in L2 (Ω, µ), where B is any set in Ω. So we only need to approximate χB (ω) with
0 < µ(B) < ∞. Since µ is a Radon measure which is inner regular, without losing
the generality, we can assume that B is compact. Then for any ω ∈ E, we can find
an x0 ∈ H such that hx0 , F(ω)i =
6 0. Let xω = x0 /hx0 , F(ω)i. Then
hxω , F(ω)i = 1.
Since F is a weakly continuous function from Ω to H, there is a neighborhood Uω
of ω such that for all υ ∈ Uω ,
r
|hxω , F(υ)i − 1| ≤
.
µ(B)
Then we have E ⊂ ∪ω∈B Uω . Since E is compact, there are finitely many {ωi }1≤i≤N
such that B ⊂ ∪1≤i≤N Uωi , and hence B = ∪1≤i≤N (Uωi ∩B). We can find a sequence
of disjoint subsets Ueωi of Uωi ∩ B such that B = ∪1≤i≤N Ueωi . Then we have
2
!
N
X
EF xω ,Ueω − χB (ω)y U
i
i
2
i=1
L (Ω,µ)
2
Z X
N
=
χUeω (ω)hxωi , F(ω)i − χB (ω) d µ(ω)
i
Ω i=1
N Z
X
2
=
|hxωi , F(ω)i − 1| d µ(ω)
i=1
≤
N
X
i=1
eω
U
i
µ(Ueωi )
= .
µ(B)
3.3. OPERATOR-VALUED MEASURES INDUCED BY CONTINUOUS FRAMES
Therefore U is surjective as claimed.
53
Then we can write the orthogonal projection-valued measure F on the dilation
space as follows:
Theorem 3.19. Let F (B) be the orthogonal projection-valued measure in Theorem 3.4. Then we have F (B) = U ∗ χB U for every B ∈ Σ.
Proof. From Theorem 3.4, we know that
!
X
X
F (B)
EF Bi ,xi =
EF B∩Bi ,xi .
i
i
Also for any y ∈ H and A ∈ Σ, we have
∗
= χB (ω)hx, F(ω)i, U EF A,y L2 (Ω,µ)
U (χB (ω)hx, F(ω)i) , EF A,y M
fE
F
D
E
=
χB (ω)hx, F(ω)i, χA (ω)hy, F(ω)i 2
L (Ω,µ)
Z
=
hx, F(ω)i · hF(ω), yid µ(ω),
B∩A
and
EF B,x , EF A,y M
f
EF
= hEF (B ∩ A)x, yiH
Z
=
hx, F(ω)i · hF(ω), yid µ(ω).
B∩A
Thus
U ∗ (χB (ω)hx, F(ω)i) = EF B,x .
Since
!
∗
U χB U
X
EF Bi ,xi
=
X
=
X
=
X
=
X
i
U ∗ χB (ω) U EF xi ,Ei
i
U ∗ χB (ω)χBi (ω)hxi , F(ω)i
i
U ∗ χB∩Bi (ω)hxi , F(ω)i
i
EF B∩Bi ,xi
i
!
= F (B)
X
EF Bi ,xi
,
i
we get F (B) = U ∗ χB U , as claimed.
Theorem 3.20. Suppose that supp F = Ω. If
L := inf{kEF (B)k : kEF (B)k > 0} > 0,
then Ω is at most countable, that is, every point in Ω is an open set.
54
3. FRAMINGS AND DILATIONS
Proof. Let L = inf{kEF (B)k : kEF (B)k > 0} > 0. First we show that for
any open subset U of Ω, we have kEF (U )k > 0. Choose ωU ∈ U and xU ∈ H such
that hxU , F(ωU )i =
6 0. Then we have
Z
2
hEF (U )xU , xU i =
|hxU , F(ω)i| d µ(ω)
U
Z
2
≥
|hxU , F(ω)i| d µ(ω)
{ω∈U :|hxU ,F (ω)i|>|hxU ,F (ωU )i|/2}
≥
µ{ω ∈ U : |hxU , F(ω)i| > |hxU , F(ωU )i|/2} · |hxU , F(ωU )i|2 /4
> 0.
Thus kEF (U )k > 0, and so kEF (U )k ≥ L for any open set U .
Fix any ω0 ∈ Ω. Since Ω is locally compact, we can choose a compact neighborhood Uω0 of ω0 . Since F(Uω0 ) is weakly compact, we have
M := sup kF(ω)k < ∞.
w∈Uω0
Thus, for any open subset U ⊂ Uω0 , we get
Z
|hEF (U )x, yi| = hx, F(ω)i · hF(ω), yid µ(ω)
ZU
≤
|hx, F(ω)i| · |hF(ω), yi|d µ(ω)
U
Z
≤
|hx, F(ω)i|2 d µ(ω)
1/2 Z
1/2
·
|hF(ω), yi|2 d µ(ω)
U
Z
≤
U
1/2 Z
1/2
kxk2 · kF(ω)k2 d µ(ω)
·
kF(ω)k2 · kyk2 d µ(ω)
U
Z
≤
U
1/2 Z
1/2
2
2
2
2
kxk · M d µ(ω)
·
M · kyk d µ(ω)
U
U
≤ µ(U ) · M 2 · kxk · kyk.
This implies that kEF (U )k ≤ M 2 µ(U ). By kEF (U )k ≥ L, we obtain that µ(U ) ≥
L/M 2 > 0 for all open subsets of Uω0 . Let
Vω0 = {V : V is an open subset of Uω0 containing ω0 }
and
e := inf µ(V ) ≥ L/M 2 > 0.
L
V ∈Vω0
e + L/(2M 2 ).
We can choose Vω0 ∈ Vω0 such that µ(Vω0 ) < L
Now, we prove that Vω0 = {ω0 }. If there is another ω1 ∈ Vω0 and ω1 6= ω0 ,
since Ω is Hausdorff, we can find an open set Wω0 ⊂ Vω0 ⊂ Uω0 containing ω0 , and
an open set Wω1 ⊂ Vω0 ⊂ Uω0 containing ω1 such that and Wω0 ∩ Wω1 = ∅. Since
e Thus,
Wω0 ∈ Vω0 , we have µ(Wω0 ) ≥ L.
e + L/(2M 2 ) > µ(Vω ) ≥ µ(Wω ) + µ(Wω ) ≥ L
e + L/M 2 ,
L
0
0
1
which is a contradiction. So {ω0 } = Vω0 is an open set, hence Ω is at most countable.
3.3. OPERATOR-VALUED MEASURES INDUCED BY CONTINUOUS FRAMES
55
Corollary 3.21. If H is separable and EF is a projection-valued measure,
then Ω is countable.
We briefly discuss one generalization of a continuous frame.
Definition 3.22. A function F : Ω → B(H, H0 ) is called an operator-valued
µ-frame if it is weakly Bochner measurable and if there exist two constants A, B > 0
such that
Z
2
AkxkH ≤
kF(ω)xk2H0 dµ(ω) ≤ Bkxk2H
Ω
holds for all x ∈ H.
Similar to the continuous frame case we have:
Theorem 3.23. Then the mapping
Z
ϕF : Σ → B(H)
hEF (B)x, yiH =
hF(ω)x, F(ω)yiH0 dµ(ω);
B
is an operator-valued measure.
Define an operator θF : H → L2 (µ; H0 ) by
(θF x)(ω) = F(ω)x,
∀ x ∈ H, ω ∈ Ω.
It is easy to see that θF is a bounded linear operator. In fact, it is injective
and bounded below. For operator-valued µ-frames the dilation space is very much
similar to the regular frames case.
Theorem 3.24. L2 (µ; H0 ) is a Hilbert dilation space of EF .
Proof. Define a mapping FF : Σ → B(L2 (µ; H0 )) by
f (ω), ω ∈ B,
(FF (B)f )(ω) =
0,
ω 6= 0.
Then FF is a self-adjoint projection-valued measure.
Since
∗
hθF
FF (B)θF x, yiH
= hFF (B)θF x, θF yiL2 (µ;H0 )
= hFF (B)F(ω)x, F(ω)yiL2 (µ;H0 )
= hχE (ω)F(ω)x, F(ω)yiL2 (µ;H0 )
Z
=
hF(ω)x, F(ω)xiH0 dµ(ω)
B
= hEF (B)x, yiH
Thus L2 (µ; H0 ) is a Hilbert dilation space of EF .
∗
The operator θF is the usual analysis operator and θF
is the usual synthesis
operator. The frame operator on H is defined by
SF : H → H,
∗
SF = θF
θF .
56
3. FRAMINGS AND DILATIONS
For any x, y ∈ H, we have
hSF x, yiH
= hθF x, θF yiL2 (µ;H0 )
Z
=
hF(ω)x, F(ω)yiH0 dµ(ω)
Ω
Z
=
hF ∗ (ω)F(ω)x, yiH dµ(ω)
Ω
Thus
Z
F ∗ (ω)F(ω)xdµ(ω).
SF x =
∗
θF
f
ZΩ
F ∗ (ω)f (ω)dµ(ω),
=
Ω
where the equation is in weak sense, i.e.
Z
∗
hθF f, xiH = hF ∗ (ω)f (ω), xiH dµ(ω),
Ω
and
Z
hSF x, yiH =
hF ∗ (ω)F(ω)x, yiH dµ(ω).
Ω
Therefore for each operator-valued µ-frame, we have four associated operatorvalued measures on (Ω, Σ), namely:
(i) The frame operator-valued measure EF : Σ → B(H) is defined by
Z
hEF (B)x, yiH =
hF(ω)x, F(ω)yiH0 dµ(ω);
B
(ii) The analysis operator-valued measure αF : Σ → B(H, L2 (µ; H0 )) is defined by
F(ω)x, ω ∈ B,
(αF (B)x)(ω) =
0,
ω 6= 0;
(iii) The synthesis operator-valued measure σF : Σ → B(L2 (µ; H0 ), H) is defined by
Z
hσF (B)f, xiH =
hf (ω), F(ω)xiH0 dµ(ω);
B
(iv) The self-adjoint projection-valued measure FF : Σ → B(L2 (µ; H0 )) is
defined by
f (ω), ω ∈ E,
(FF (E)f )(ω) =
0,
ω 6= 0.
CHAPTER 4
Dilations of Maps
In this chapter we establish some dilation results for general linear mappings of
algebras, mainly focusing on (not necessarily cb-maps) on von Neumann algebras,
and more generally on Banach algebras. The ideas for our proofs come indirectly
from our methods in Chapter 2 for OVM’s.
We begin with a possibly known purely algebraic result (Proposition 4.1) which
shows that dilations of linear maps are always possible even in the absence of any
topological structure. In the presence of additional hypotheses stronger results are
possible: When a domain algebra, mapping and range space have strong continuity
and/or structural properties we seek similar properties for the dilation. This plan
led to our other results. Theorem 4.10 states that for any Banach algebra A and
any bounded linear operator φ from A to B(H) on a Banach space H, there exist
a Banach space Z, a bounded linear unital homomorphism π : A → B(Z), and
bounded linear operators T : H → Z and S : Z → H such that
φ(a) = Sπ(a)T
for all a ∈ A . In the case that the Banach algebra is an abelian purely atomic von
Neumann algebra A and H is a Hilbert space, and φ is normal (i.e. ultraweakly
continuous), then there is a normal dilation π (Theorem 4.7). If φ is not cb then
the dilation space cannot be a Hilbert space; the normality of the dilation is with
respect to the natural ultraweak topology on B(Z). It is not known the extent to
which this result can be generalized (i.e. achieving normality of the dilation).
4.1. Algebraic Dilations
Proposition 4.1. If A is unital algebra, V a vector space, and φ : A → L(V )
a linear map, then there exists a vector space W , a unital homomorphism π : A →
L(V ), and linear maps T : V → W , S : W → V , such that
φ(·) = Sπ(·)T.
Proof. For a ∈ A, x ∈ V , define αa,x ∈ L(A, V ) by
αa,x (·) := φ(·a)x.
Let W := span{αa,x : a ∈ A, x ∈ V } ⊂ L(A, V ). Define π : A → L(W ) by
π(a)(·) := α·a,x . It is easy to see that ρ is a unital homomorphism. For x ∈ V
define T : V → L(A, V ) by Tx := αI,x = φ(·I)x = φ(·)x. Define S : W → W
by setting S(αa,x ) := φ(a)x and extending linearly to W . If a ∈ A, x ∈ V are
arbitrary, we have Sπ(a)T x = Sπ(a)αI,x = Sαa,x = φ(a)x. Hence φ = SπT.
57
58
4. DILATIONS OF MAPS
4.2. The Commutative Case
We first examine a mapping from the commutative C ∗ -algebra `∞ into B(H)
which is induced by a framing on a Hilbert space H. Here `∞ means with respect
to a countable or finite index set J, and it is well known that every separably
acting purely atomic abelian von Neumann algebra is equivalent to some `∞ via an
ultraweakly continuous ∗-isomorphism.
Theorem 4.2. Let H be a separable Hilbert space and let (xi , yi ) be a framing
of H. Then the mapping φ from `∞ into B(H) defined by
XSOT
ai xi ⊗ yi
φ : `∞ → B(H),
(ai ) →
is well-defined, unital, linear and ultraweakly continuous.
Proof. Since (xi , yi ) is a framing of H, for any x ∈ H,
X
x=
hx, yi ixi
i
converges unconditionally.
For any (ai ) ∈ c00 , define a bounded operator U(ai ) as follows
X
U(ai ) : H → H, U(ai ) (x) =
ai hx, yi i xi .
P
For any x ∈ H, since i hx, yi ixi converges unconditionally, by [DJT] Theorem
1.9, we know that
X
bi hx, yi i xi < +∞.
sup
(bi )∈B1 (`∞ )
Thus,
U(a ) (x) ≤
i
sup
(ai )∈B1 (c00 )
sup
(bi )∈B1 (`∞ )
X
bi hx, yi i xi < +∞.
Then by the Uniform Boundedness Principle,
U(a ) < +∞.
sup
i
(ai )∈B1 (c00 )
It follows that
Ku =
sup
X
sup σi hx, yi i xi < +∞.
x∈B1 (H) (σi )⊂D
Thus, for all (ai ) ∈ `∞ and x ∈ H,
X
ai hx, yi i xi ≤ Ku kai k`∞ kxk.
Hence F is well-defined, unital, linear and bounded with
kF kB(`∞ ,B(H)) ≤ Ku .
Now we prove that F is ultraweakly continuous. If there
a net (aλi ) converges to
P is
λ
0 in the ultraweakly topology, then for any (γi ) ∈ `1 ,
ai γi → 0. Let T belong to
the trace class S1 (H). By the polar decomposition, T = U |T | where U is a partial
isometry. Moreover, recall that S1 (H) is the subset of the compact operators K(H),
|T | is a self-adjoint compact operator. Thus there is an orthonormal basis (ei ) and
a sequence λi ≥ 0 so that
X
|T | =
λi ei ⊗ ei
4.2. THE COMMUTATIVE CASE
59
P
with kT kS1 = tr(|T |) =
i λi < ∞. Then for all (ai ) ∈ `∞ , (γj ) ∈ `1 and
(uj ), (vj ) ⊂ B1 (H), we have
XX
X
X
|ai γj huj , yi i hxi , vj i| ≤ k(ai )k∞
|γj |
|huj , yi i hxi , vj i|
i
j
= k(ai )k∞
j
i
X
X
|γj |
j
= k(ai )k∞
X
|γj |
j
≤
k(ai )k∞
X
k(ai )k∞
X
j
≤
θi,j huj , yi i hxi , vj i
i
*
X
+
θi,j huj , yi i xi , vj
i
X
θi,j huj , yi i xi |γj | sup j i
|γj |Ku < ∞,
j
where θi,j = sgn {huj , yi i hxi , vj i}. So we have
XX
aλi λj hU ej , yi i hxi , ej i < ∞
i
j
and
X X
λ
hU
e
,
y
i
hx
,
e
i
j
j i
i j < ∞.
i j
Therefore
tr φ(aλi )T
=
X
=
X
φ(aλi )T ej , ej
j
*
!
φ(aλi )U
j
X
λk ek ⊗ ek
+
ej , ej
k
=
X
=
X
=
X
=
X
φ(aλi )U λj ej , ej
j
λj
*
X
j
aλi xi ⊗ yi U ej , ej
i
λj
X
aλi
X
j
i
+
aλi hU ej , yi i hxi , ej i ,
i
λj hU ej , yi i hxi , ej i ,
j
which converges to 0, as claimed.
The main purpose of this section is to show that for every ultraweakly continuous mapping φ from a purely atomic abelian von Neumann algebra A into B(H),
we can find a Banach space Z, an ultraweakly continuous unital homomorphism π
from A into B(Z), and bounded linear operators T and S such that for all a ∈ A,
φ(a) = Sπ(a)T.
60
4. DILATIONS OF MAPS
This result differs from Stinespring’s dilation because the map φ here is not necessarily completely bounded and consequently the dilation space is not necessarily a
Hilbert space.
While the ultraweak topology on B(H) for a Hilbert space H is well-understood,
we define the ultraweak topology on B(X) for a Banach space X through tensor
products: Let X ⊗ Y be the tensor product of the Banach space X and Y. The
projective norm on X ⊗ Y is defined by:
( n
)
n
X
X
kuk∧ = inf
kxi kkyi k : u =
xi ⊗ yi .
i=1
i=1
We will use X ⊗∧ Y to denote the tensor product X ⊗Y endowed with the projective
b From [R] Section 2.2, for any
norm k · k∧ . Its completion will be denoted by X ⊗Y.
Banach spaces X and Y, we have the identification:
b )∗ = B(X, Y ∗ ).
(X ⊗Y
b ∗ )∗ . Viewing X ⊆ X ∗∗ , we define the ultraweak topology
Thus B(X, X ∗∗ ) = (X ⊗X
b ∗ . We will usually
on B(X) to be the weak* topology induced by the predual X ⊗X
use the term normal to denote an ultraweakly continuous linear map.
The following lemma generalizes Theorem 4.2 and will be used in the proof of
Theorem 4.4 of this section. The proof is similar to that of Theorem 4.2 and we
include a sketch for completeness.
Lemma 4.3. Let X be a Banach space and let E : 2N → B(X) be an operatorvalued measure on (N, 2N ). Denote E({i}) by Ei for all i ∈ N. Then the mapping
φ from `∞ into B(X) defined by
XSOT
φ : `∞ → B(X),
(ai ) 7→
ai Ei
is well-defined, linear and ultraweakly continuous.
Proof. Since E : 2N → B(X) is an operator-valued measure on X, we have
for all x ∈ X,
X
Ei (x)
i
converges unconditionally. Similar to the proof in Theorem 4.2, we get that
X
Ku = sup
sup σi Ei (x) < +∞,
x∈B1 (X) (σi )⊂D i
and so for all (ai ) ∈ `∞ and x ∈ X, we have
X
ai Ei (x) ≤ Ku kai k`∞ kxk.
i
Thus φ is well-defined, linear and bounded with
kφkB(`∞ ,B(X)) ≤ Ku .
For the ultraweakly continuity of φ, let P
(aλi ) be a net converging to 0 in the
ˆ π X ∗ . Then
ultraweak topology. Then for any (γi ) ∈ `1 ,
aλi γi → 0. Let w ∈ X ⊗
4.2. THE COMMUTATIVE CASE
61
∗
there
is a pair of sequences
P
P(uj , vj ) ⊂ X/{0} × X /{0} with the property that
kuj kkvj k < ∞ and w =
uj ⊗ vj . Thus for all (ai ) ∈ `∞ , we have
XX
X
X
|ai hEi (uj ), vj i| =
|ai |
|hEi (uj ), vj i|
i
j
i
≤
j
k(ai )k∞
XX
i
= k(ai )k∞
|hEi (uj ), vj i|
j
XX
j
|hEi (uj ), vj i|
i
= k(ai )k∞
X
= k(ai )k∞
X
kuj kkvj k
X uj vj Ei
,
kuj k
kvj k kuj kkvj k
X
j
i
j
θi,j
Ei
i
uj
kuj k
,
vj
kvj k
vj
,
Ei
here θi,j = sgn
kvj k
*
+
X
X
uj
vj
= k(ai )k∞
kuj kkvj k
θi,j Ei
,
kuj k
kvj k
j
i
X
X
uj ≤ k(ai )k∞
kuj kkvj k sup θi,j Ei
ku
k
j j
j
i
X
≤ k(ai )k∞
kuj kkvj kKu < ∞.
uj
kuj k
j
Thus we get that
XX
XX
X
X
|aλi hEi (uj ), vj i | =
|aλi hEi (uj ), vj i | =
|aλi |
| hEi (uj ), vj i | < ∞
j
i
and
i
j
i
j
XX
X X
≤
hE
(u
),
v
i
| hEi (uj ), vj i | < ∞.
i
j
j
i j
i
j
So


X

hEi (uj ), vj i ∈ `1 .
j
i
Therefore
φ(aλi )w
=
X
φ(aλi )uj , vj
j
=
*
X X
j
i
=
XX
=
X
j
i
+
aλi Ei (uj ), vj
aλi hEi (uj ), vj i
i
aλi
X
j
hEi (uj ), vj i
62
4. DILATIONS OF MAPS
converges to 0, as expected.
Theorem 4.4. Let H be a separable Hilbert space and φ : `∞ (N) → B(H) such
that φ(1) = I and φ(en ) is at most rank one for all n ∈ N, where en = χ{n} and 1
is the function 1 in `∞ . Then the following are equivalent:
(i) φ is ultraweakly continuous,
P
(ii) the induced measure E defined by defined by E(B) = φ( n∈B en ) for any
B ⊆ N is an operator-valued measure;
(iii) φ is induced by a framing (xn , yn ) for H, i.e,
X
XSOT
φ(
an en ) =
an xn ⊗ yn
n∈N
Proof. (i) ⇒ (ii) and (i) ⇒ (iii) are obvious from the definition of an operator
valued measure and the ultraweakly continuity of φ. (ii) ⇒ (i) follows from Lemma
4.3. (iii) ⇒ (i) follows from Theorem 4.2.
Corollary 4.5. Let H be a separable Hilbert space and φ : `∞ → B(H)
such that φ(1) = I and φ(en ) is at most rank one for all n ∈ N. Then there
exist a separable Banach space Z, an ultraweakly continuous unital homomorphism
π : `∞ → B(Z), and bounded linear operators T : H → Z and S : Z → H such
that
φ(a) = Sπ(a)T
for all a ∈ `∞ , and π(en ) is rank one for all n ∈ N.
Proof. By Theorem 4.4, φ is induced by a framing (xn , yn ) for H. Thus by
Theorem 4.6 in [CHL] (xn , yn ) can be dilated an unconditional basis {un } for a
Banach space Z (Hence Z is separable). Let π be the induced operator valued
map by (uλ , u∗λ ) (where {u∗λ } is the dual basis of {uλ }). Then π satisfies all the
requirements.
By using our main dilation result in Chapter 2 we are able to generalize the
above result to more general ultraweakly continuous operator valued mapping.
Theorem 4.6. Let φ : `∞ → B(H) be an ultraweakly continuous linear mapping. Then there exists a Banach space Z, an ultraweakly continuous unital homomorphism π : `∞ → B(Z), and bounded linear operators T : H → Z and S : Z → H
such that
φ(a) = Sπ(a)T
for all a ∈ `∞ .
Proof. Let E : 2N → B(H) be defined by
E(N ) = φ(χN )
for all N ⊂ N.
If (Ni ) is a sequence of disjoint subsets of N with union N , then it follows easily that
P
χNi converges to χN under the ultraweak topology of `∞ . Since φ is ultraweakly
continuous and x ⊗ y belongs to the trace class S1 (H) for all x, y ∈ H,we get that
=
hE(N )x, yi = hφ(χN )x, yi = φ(χN )(x ⊗ y)
X
X
X
φ(χNi )(x ⊗ y) =
hφ(χNi )x, yi =
hE(Ni )x, yi.
i
i
i
4.2. THE COMMUTATIVE CASE
63
Thus E : 2N → B(H) is an operator-valued measure on (N, 2N ). Let
fE,M , ρM , SM , TM )
(Ω, Σ, M
fE )
be its minimal dilation system. By Lemma 4.3, the mapping π from `∞ to B(M
defined by
XSOT
fE,M ), π(ai ) =
π : `∞ → B(M
ai ρM ({i})
i
is an ultraweakly continuous unital homomorphism. Moreover, for all (ai ) ∈ `∞
and x ∈ H, we have
X
X
X
SM π(ai )TM x = SM
ai ρM ({i})Ex,N = SM
ai Ex,{i} =
ai E({i})x = φ(ai )x.
i
i
i
This completes the proof.
Since every separably acting purely atomic abelian von Neumann algebra is
equivalent to some `∞ via an ultraweakly continuous ∗-isomorphism, we immediately get the following:
Theorem 4.7. Let A be a purely atomic abelian von Neumann algebra acting
on a separable Hilbert space. Then for every ultraweakly continuous linear map
φ : A → B(H), there exists a Banach space Z, an ultraweakly continuous unital
homomorphism π : `∞ → B(Z), and bounded linear operators T : H → Z and
S : Z → H such that
φ(a) = Sπ(a)T
for all a ∈ A.
Every ultraweakly continuous linear map φ : L∞ (Ω, Σ, µ) → B(H) induces an
OVM E : (Ω, Σ) → B(H). Please notice that here E is absolutely continuous with
respect to µ, E µ, that is, µ(E) = 0 implies E(E) = 0.
Lemma 4.8. If E : (Ω, Σ) → B(H) is an OVM, and if µ is a non-negative
scalar measure on (Ω, Σ) such that E µ (that is, E is absolutely continuous
with respect to µ), then there exists a bounded linear map φ : L∞ (µ) → B(H) that
induces E on (Ω, Σ), (that is, E(B) = φ(χB ) for all B ∈ Σ).
Pn
For any
simple function P i=1 αi χBi with
P
Pmdisjoint Bi ∈ Σ, define
PProof.
n
n
n
α
E(B
).
If
α
χ
=
φ( i=1 αi χBi ) =
i
i
i
B
i
i=1
i=1
j=1 βj χAj for disjoint Bi
Pn Pm
and disjoint Aj , then i=1 j=1 (αi − βj )χBi ∩Aj = 0. Thus, if αi − βj 6= 0, then
Pm
Pn
µ(Bi ∩ Aj ) = 0. By E µ, we obtain that φ( i=1 αi χBi ) = φ( j=1 βj χAj ), φ is
well-defined on the subspace of all simple
functions
in L∞ (µ). ToPprove
that φ is
Pn
Pm
n Pm
linear, we only need to notice that i=1 αi χBi + j=1 βj χAj = i=1 j=1 (αi +
βj )χBi ∩Aj . For the boundedness, we need the following uniformly boundedness
first, since supE∈Σ kE(B)k < ∞, it is easy to obtain that
{Bi }n
i=1
sup
is a partition of
sup
n
X
Ω kxk,kyk≤1 i=1
|hE(Bi )x, yi| = Mφ < ∞.
64
4. DILATIONS OF MAPS
Then we have
n
!
n
X
X
αi E(Bi )
αi χBi = φ
i=1
i=1
=
≤
sup
n
X
|αi hE(Bi )x, yi|
kxk,kyk≤1 i=1
n
X
sup
!
|hE(Bi )x, yi|
kxk,kyk≤1 i=1
n
X
≤ Mφ i=1
αi χBi max |αi |
1≤i≤n
.
L∞
Because the simple functions are dense in L∞ (µ), the conclusion follows.
For an OVM E : (Ω, Σ) → B(H), where H is a separable Hilbert space, there
always exists a non-negative scalar measure µ on (Ω, Σ) such that E µ holds. In
fact, assume that {xn } is dense in the unit sphere of H, SH , take
X
µ(B) =
αnm |hE(B)xn , xm i|,
n,m
where αnm > 0 and
P
n,m
αnm = 1. Then we immediately get that
Corollary 4.9. If E : (Ω, Σ) → B(H) is an OVM, then there exists µ which
is a non-negative scalar measure on (Ω, Σ) such that E µ, and a bounded linear
map φ : L∞ (µ) → B(H) that induces E on (Ω, Σ).
We remark that the conclusion of Corollary 4.9 actually holds for more general
von Neumann algebras [BW]: Let A be a von Neumann algebra without direct
summand of type I2 . Then every bounded and finitely additive B(H)-valued measure (c.f. [BW] for definition) on the projection lattice of A can be uniquely extend
to bounded linear map from A to B(H). We include Lemma 4.8 and Corollary 4.9
here since they are directly related to the following problem. We think if the answer
to Problem C is positive then a solution might be along the lines of the proof of
Lemma 4.8.
Problem C. Let E : (Ω, Σ) → B(H) be an OVM. Is there an ultraweakly continuous map φ : L∞ (µ) → B(H) that induces E on (Ω, Σ)?
4.3. The Noncommutative Case
Theorem 4.10. [Banach Algebra Dilation Theorem] Let A be a Banach
algebra, and let φ : A → B(H) be a bounded linear operator, where H is a Banach
space. Then there exists a Banach space Z, a bounded linear unital homomorphism
π : A → B(Z), and bounded linear operators T : H → Z and S : Z → H such that
φ(a) = Sπ(a)T
for all a ∈ A .
4.3. THE NONCOMMUTATIVE CASE
65
Proof. Consider the algebraic tensor product space A ⊗ H, and define a
B(A , H) operator tensor norm with respect to φ as follows: for any a ∈ A and
x ∈ H, identity a ⊗ x with the map a → H defined by
(a ⊗ x)(b) = φ(ba)x.
Then a ⊗ x is a bounded linear operator from A to H with ka ⊗ xkB(A ,H) ≤
kφkkakkxk. So this defines a quasi-norm on A ⊗ H. Let
(
N =
)
X
ci ai ⊗ xi : ||
X
i
ci ai ⊗ xi ||B(A ,H) = 0
i
e φ H/N be the completion of the norm space A ⊗ H/N .
and A ⊗
For any a ∈ A , let us define π(a) : A ⊗ H/N → A ⊗ H/N by
!
X
π(a)
ai ⊗ xi
=
X
(aai ) ⊗ xi .
i
Assume that f =
Pn
i=1
i
ai ⊗ xi =
Pm
j=1 bj
⊗ yi . Then we have
!
ρ(a)
X
ai ⊗ xi
!
(b)
=
X
i
(aai ) ⊗ xi
(b)
i
=
X
φ(baai )xi
i
!
=
X
ai ⊗ xi
(ba)
i
= f (ba),
and
!
ρ(a)
X
bj ⊗ yj
!
(b)
=
X
i
(abj ) ⊗ yj
(b)
i
=
X
φ(babj )yj
i
!
=
X
bj ⊗ y j
i
= f (ba).
Therefor π(a) is well defined.
(ba)
66
4. DILATIONS OF MAPS
The boundedness π(a) (with kπ(a)k ≤ kak) follows from
!
X
ai ⊗ xi kπ(a)(f )k = π(a)
i
X
= (aai ) ⊗ xi i
! X
= sup (aai ) ⊗ xi (b)
b∈BA
i
X
φ(baai )xi = sup b∈BA i
X
a
φ(b
= kak sup ai )xi kak
b∈BA
i
X
φ(bai )xi ≤ kak sup b∈BA i
! X
= kak sup ai ⊗ xi (b)
b∈BA
i
X
= kak ai ⊗ xi i
= kak · kf k.
e φ H/N which we still denote
Extend π(a) to a bounded linear operator on A ⊗
e φ H/N is a bounded linear operator. Moreover,
it by π(a). Thus π : A → A ⊗
!
X
X
π(ab)
ai ⊗ xi
=
(abai ) ⊗ xi
i
i
!
=
π(a)
X
(bai ) ⊗ xi
i
!
=
π(a)π(b)
X
ai ⊗ xi
i
Hence
π(ab) = π(a)π(b),
and therefore π is a homomorphism.
e φ H/N by T (x) = 1 ⊗ x. Since
Define T : H → A ⊗
kT (x)k
= k1 ⊗ xk
=
sup k(1 ⊗ x) (a)k
a∈BA
=
sup kφ(a)xk
a∈BA
≤
sup kakkφkkxk
a∈BA
= kφkkxk,
.
4.3. THE NONCOMMUTATIVE CASE
67
we have that T is a bounded linear operator with kT k ≤ kφk.
Define S : A ⊗ H/N → H by S(a ⊗ x) = E(a)x and linearly extend S to
A ⊗ H/N . It is easy to check that S is well-defined. Since
kS(a ⊗ x)k
kφ(a)xk
=
≤
sup kφ(ba)xk
b∈BA
sup k(a ⊗ x) (b)k
=
b∈BA
ka ⊗ xk,
=
we have that S is a bounded linear operator with kSk ≤ 1. Extend S to a bounded
e φ H/N to H, which we still denote it by S.
linear operator from A ⊗
Finally, for any x ∈ H we have
Sπ(a)T (x)
=
Sρ(a) (1 ⊗ x)
=
S (a ⊗ x)
=
φ(a)x.
Thus φ(a) = Sπ(a)T.
Remark 4.11. Consider the construction of the dilation space in Theorem 4.10,
and compare it with the construction of the dilation space in Theorem 4.7. In the
first case we take the algebraic tensor product A ⊗H considered as a linear space of
operators in B(A , H) to obtain a quasi-norm on A ⊗H which we denote k·kB(A ,H) .
Then we mod out by the kernel to obtain a norm on the quotient space, and then
we complete it to obtain a Banach space. If we began this construction by using
a dense subsalgebra of A instead of A itself, then the construction goes through
smoothly, and the dilation space is the same. In the case of the second space we
work with A = `∞ , and the reader can notePthat we start with the algebraic
tensor product of the dense subalgebra A0 = { ci χEi : ci ∈ C, Ei ∈ Σ} with H,
and define a special “minimal” quasi-norm k · k1 on it as in Section 2.4, then mod
out by the kernel, and complete it to obtain the dilation space. In both cases we
could begin with the dense subalgebra A0 of `∞ , form the algebraic tensor product
A0 ⊗ `∞ , and put a quasi-norm on the space. It is easy to show that these two
quasi-norms are equivalent in the sense that each one is dominated by a constant
multiple of the other. Hence the dilation spaces of Theorem 4.7 and Theorem 4.10
are equivalent for the special case A = `∞ .
As with Stinespring’s dilation theorem, if A and H in Theorem 4.10 are both
separable then the dilated Banach space Z is also separable. However, the Banach
algebras we are interested in include von Neumann algebras and these are generally
not separable, and the linear maps φ : A → B(H) are often normal. So we pose
the following two problems.
Problem D. Let K, H be separable Hilbert spaces, let A ⊂ B(K) be a von Neumann algebra, and let φ : A → B(H) be a bounded linear map. When is there a
separable Banach space Z, a bounded linear unital homomorphism π : A → B(Z),
and bounded linear operators T : H → Z and S : Z → H such that
φ(a) = Sπ(a)T
for all a ∈ A ?
68
4. DILATIONS OF MAPS
Problem E. Let A ⊂ B(K) be a von Neumann algebra, and φ : A → B(H) be a
normal linear map. When can we dilate φ to a normal linear unital homomorphism
π : A → B(Z) for some Banach space Z?
Although we do not know the answer to Problem E, we do have the following
result:
Theorem 4.12. Let K, H be Hilbert spaces, A ⊂ B(K) be a von Neumann
algebra, and φ : A → B(H) be a bounded linear operator which is ultraweakly-SOT
continuous on the unit ball BA of A. Then there exists a Banach space Z, a bounded
linear homeomorphism π : A → B(Z) which is SOT-SOT continuous on BA , and
bounded linear operator T : H → Z and S : Z → H such that
φ(a) = Sπ(a)T
for all a ∈ A.
Proof. We only need to prove that the bounded linear homeomorphism π
constructed in Theorem 4.10 is SOT-SOT continuous.
Assume that a net {aλ } ⊂ BA . It is sufficient to prove that aλ → 0 (SOT)
implies that π(aλ ) → 0 (SOT). Since {π(aλ )} is norm-bounded, we have that
π(aλ ) → 0 (SOT) if and only if π(aλ )ξ → 0 in norm for a dense set of ξ. Thus, we
only need to prove that
π(aλ )(a ⊗ x) = (aλ a) ⊗ x → 0
in the norm topology for all a ∈ A and x ∈ H. If this is not the case, then there
exist δ > 0, a subnet of {aλ } (still denoted by {aλ }), and {bλ } ⊂ BA such that
k(aλ a ⊗ x)(bλ )k = kϕ(bλ aλ a)(x)k > δ.
Since on any norm-bounded set the WOT and ultraweak topology are the same
(and in particular the unit ball is compact in both topologies), there is a subnet
of {bλ } (still denoted by {bλ }) converges to some element b ∈ BA in the ultraweak
topology (or WOT). Thus, from our hypothesis that aλ → 0 (SOT), we get that
bλ aλ a → 0 (SOT). This implies that bλ aλ a → 0 in ultraweak topology (or WOT),
and therefore ϕ(bλ aλ a) → 0 (SOT) since ultraweakly-SOT continuous on BA . This
leads to a contradiction.
Corollary 4.13. Let K, H be separable Hilbert spaces, A ⊂ B(K) be a von
Neumann algebra, and φ : A → B(H) be a bounded linear operator which is
ultraweakly-SOT continuous on BA . Then there exists a separable Banach space
e a bounded linear homeomorphism π̃ : A → B(Z)
e which is SOT-SOT continuous
Z,
e
e
e → H such that
on BA , and bounded linear operator T : H → Z and Se : Z
φ(a) = Seπ̃(a)Te
for all a ∈ A.
Proof. By Theorem 4.12, there exists a Banach space Z, a bounded linear homeomorphism π : A → B(Z) which is SOT-SOT continuous on BA , and
bounded linear operator T : H → Z and S : Z → H such that φ(a) = Sπ(a)T for
all a ∈ A. Since K is separable, there is {ai }∞
i=1 SOT dense in BA . Define
e = spank·k {π(ai )T (h) : 1 ≤ i < ∞, h ∈ H}.
Z
4.3. THE NONCOMMUTATIVE CASE
69
e is separable since H is separable. Moreover, the SOT-SOT continuity of π
Then Z
e Let
(restricted to the unit ball of A) implies that π(A)Ze ⊂ Ze and π(A)T (H) ⊂ Z.
V : Ze → Z be the inclusion linear map. Then
φ(a) = SV −1 π(a)V T, ∀ a ∈ A.
e Then π̃ is a dilation of φ on the separable
Let π̃(a) = V −1 π(a)V acting on Z.
e is also SOT-SOT continuous when restricted
Banach space Ze and π̃ : A → B(Z)
to the unit ball of A.
Remark 4.14. Dilations and the similarity problem: Let A be a C ∗ -algebra. In
1955, Kadison [K1] formulated the following still open conjecture: Any bounded
homomorphism π from a C ∗ -algebra A into the algebra B(H) of all bounded
operators on a Hilbert space H is similar to a ∗-homomorphism, i.e. there is
an invertible operator T ∈ B(H) such that T π(·)T −1 is a ∗-homomorphism from
A to B(H). This problem is known to be equivalent to several famous open
problems (c.f. [Pi]) including the derivation problem: Is every derivation from
a C ∗ -algebra A ⊆ B(H) into B(H) inner? While Kadison’s similarity problem
remains unsettled, many remarkable partial results are known. In particular, it is
well known that π is similar to a ∗-homomorphism if and only if it is completely
bounded. It has been proved that a bounded unital homomorphism π : A → B(H)
is completely bounded (and hence similar to a ∗-representation) if A is nuclear
[Bun, Chr]; or if A = B(H), or more generally if A has no tracial states; or if A
is commutative; or if A is a II1 -factor with Murry and von Neumann’s property Γ;
or if π is cyclic [Haa]. Therefore if A belongs to any of the above mentioned classes,
and φ : A → B(H) is a bounded but not completely bounded linear map, then the
dilation space Z in Theorem 4.10 [Banach algebra dilation theorem] can never be
a Hilbert space since otherwise π : A → B(Z) would be completely bounded and
so would be φ. On the other hand, if there is a non completely bounded map φ
from a C ∗ -algebra to B(H) that has a Hilbert space dilation: π : A → B(Z) (i.e.,
where Z is a Hilbert space), then it would be a counterexample to the Kadison’s
similarity problem. So we have the following question: Is there a non-cb map that
admits a Hilbert space dilation to a bounded homomorphism?
Remark 4.15. For a countable index set Λ, there is a 1-1 correspondence between the set of (discrete) framings on a Hilbert space H indexed by Λ and the set
of ultraweakly continous unital linear maps from `∞ (Λ) into B(H). Here, unital
means it takes the function 1 in `∞ (Λ) to the identity operator in B(H). There
is also a 1-1 correspondence between the set of (discrete) framings on a Hilbert
space H indexed by Λ and the set of purely atomic probability operator-valued
measures on the σ-algebra of all subsets of Λ with rank-1 atoms in B(H). So we
have a space of ordered triples which consists of a discrete framing, a purely atomic
operator-valued probability measure with rank-1 atoms, and an ultraweakly continous unital linear map from an purely atomic abelian von Neumann algebra into
B(H) where each minimal projection is sent to a rank ≤ 1 operator. Each item in
a triple determines the other items uniquely. There is a consistent dilation theory,
where the dilation of the discrete framing, the operator-valued measure, and the
ultraweakly continous unital linear map, all have the same dilation space and the
dilation procedure commutes with the correspondences in the triple, i.e. when we
go from framing to OVM to linear map and then dilate each one to get a dilated
70
4. DILATIONS OF MAPS
triple, it will be the same as if we dilate any one and then derived the other two
from it in the natural way. Similarly for a more general index set Λ (i.e. compact
or locally compact, whatever is more suitable) there are connections among the set
of operator-valued probability measures on Λ taking values in B(H) for a Hilbert
space H, the set of unital ultraweakly continuous maps from `∞ (Λ) into B(H), and
their corresponding dilation theory. Although not every ultraweakly continuous
unital linear mapping is associated with a (continuous) framing, there are many
ultraweakly continuous unital linear mappings that are induced by framings. For
this reason we can view ultraweakly continuous unital linear mappings from `∞ (Λ)
to B(H) as abstract framings, which is the commutative theory because the domain
of the map is a commutative von neumann algebra. When we pass to the noncommutative domains, the dilation theory for various linear maps can be viewed as a
noncommutative (abstract) framing dilation theory.
CHAPTER 5
Examples
We provide the details of the two examples mentioned in the introduction
chapter. Our first example shows that there exists a framing for a Hilbert space
whose induced operator valued measure fails to admit a Hilbert space dilation.
Equivalently, it cannot be re-scaled to obtain a framing that admits a Hilbert space
dilation. The construction is based on an example of Ozaka [Os] of a normal
non-completely bounded map of `∞ (N) into B(H).
Theorem 5.1. There exist a framing for a Hilbert space such that its induced
operator-valued measure is not completely bounded, and consequently it can not be
re-scaled to obtain a framing that admits a Hilbert space dilation.
Since if a framing admits a Hilbert space dilation, then the induced operator
valued map is completely bounded and a re-scaled framing induces the same operator valued map, we only need to show that there exists a framing for a Hilbert
space such that its induced operator-valued measure is not completely bounded.
We need the following lemmas.
Lemma 5.2. Let {An } be a sequence of finite-rank bounded linear operators on
a Hilbert space H such that
(i) An Am = Am An = 0 for all n 6= m;
(ii) there exist mutually orthogonal projections {Pn } such that An = Pn An Pn
for all n.P
∞
(iii) n=1 An converges unconditionally to A ∈ B(H) with the strong operator
topology.
Assume that An has the rank one operator decomposition An = Qn,1
P+...+Qn,kn
such that for any subset Λ of Jn := {(n, 1), ..., (n, kn )} we have || j∈Λ Qj || ≤
P
||An ||. Let J be the disjoint union of Jn (n = 1, 2, ...). Then the series j∈J Qj
converges unconditionally to A.
Proof.PLet x ∈ H. Without losing the generality we can assume that x ∈ P H
∞
where P = n=1 Pn . Let > 0. Then there exists N such that
||
N
X
Pn x − x|| < /2||A||.
n=1
Now let {j` }∞
and let L be such that {j1 , ...., jL } contain
`=1 be an enumeration of J P
PN
N
⊥
∪N
J
.
Write
x
=
x
+
x
with
x
=
1
0
1
n=1 n
n=1 Pn x and x0 = (
n=1 Pn ) x. Then
0
for any L ≥ L, we get
0
L
X
`=1
Qj` x =
N X
X
n=1 j` ∈Jn
Qj` x +
X
Qj` x =
j` ∈J
/ n ,`≤L0
71
N
X
n=1
An x +
X
j` ∈J
/ n ,`≤L0
Qj` x0
72
5. EXAMPLES
P
where we use the property that j` ∈J
/ n ,`≤L0 Qj` x1 = 0. Moreover, from our assumptions on {An } and their rank-one decompositions we also have
X
Qj` x0 || ≤ sup{||An } ≤ ||A||
||
j` ∈J
/ n ,`≤L0
and
N
X
An x = A
n=1
N
X
Pn x.
n=1
Therefor we get
0
||
L
X
Qj` x − Ax||
≤
||A
N
X
≤
Qj` x0 ||
j` ∈J
/ n ,`≤L0
n=1
`=1
X
Pn x − Ax|| + ||
||A|| · ||
N
X
Pn x − x|| + ||A|| · ||x0 ||
n=1
=
2||A|| · ||x0 || < .
This completes the proof.
Lemma 5.3. Let A be a rank-k operator. Then there exists rank-one decompoPk
P
sition A = i=1 Qi such that || i∈I Qi || ≤ ||A|| for any subset I of {1, 2, ..., k}.
Pk
Proof. Let A = U |A| be its polar decomposition and write |A| = i=1 xi ⊗xi .
P
Pk
Then we have || i∈I xi ⊗ xi || ≤ || i=1 xi ⊗ xi || = ||A||. Let Qi = U xi ⊗ xi . Then
we get
X
X
||
Qi || = ||U
xi ⊗ xi ||
i∈I
i∈I
≤
||U || · ||
X
xi ⊗ xi ||
i∈I
≤
||
X
xi ⊗ xi ||
i∈I
≤
||
k
X
xi ⊗ xi || = ||A||
i=1
holds for any subset I of {1, 2, ..., k}, as claimed.
Now we prove Theorem 5.1: By Lemma 2.4 (or Lemma 2.1) in [Os] there
∞
∞
exists a σ-weakly continuous bounded linear map ψ from ⊕∞
n=1 `n into ⊕n=1 M2n
2n
(as a subalgebra acting on H := ⊕C ) which is not completely bounded and
∞
n
n
ψ = ⊕∞
n , where ψn : `n → M2n and M2n is the algebra of 2 × 2 -matrices.
n=1 ψP
n
n
n ∞
n
Let Bn =
i=1 ψ(ei ), where ei `n such that ei (j) = δi,j , and let B = ψ(e)
where e ∈ ⊕∞
`∞
n is unital element. Let Pn be the orthogonal projections from
n=1
2n
H onto C . Without losing the generality we can assume that ||ψ|| < 1. Let
An = Bn + Pn . Then {An } satisfies all the conditions (i)–(iii) in Lemma 5.6, and
P
∞
n=1 An converges to A = B + I unconditionally in the strong operator topology.
By Lemma 5.7, we can decompose each An = Qn,1
P+ ... + Qn,kn such that for
any subset Λ of Jn := {(n, 1), ..., (n, kn )} we have || j∈Λ Qj || ≤ ||An ||. Thus, by
P
Lemma 5.6, j∈J Qj converges unconditionally to A. Write Qj = xj ⊗ yj , and
5. EXAMPLES
73
P
let uj = A−1 xj . Then {uj , yj } form a framing for H since j∈J uj ⊗ yj converges
unconditionally to I in the strong operator topology.
We claim that the induced operator-valued map Φ : `∞ (J) → B(H) is not
completely bounded. In fact, if it is completely bounded, then operator valued
∞
map Ψ by {xj , yj } will be completely bounded. By embedding ⊕∞
n=1 `n naturally
∞
∞
∞
∞
to ` (J), we can view ⊕n=1 `n as a subalgebra of ` (J), and so the restriction
∞
of Ψ to this subalgebra is ψ + id, where id is the natural embedding of ⊕∞
n=1 `n
into B(H). This ψ + id is completely bounded and so ψ is completely bounded
since the embedding map id is completely bounded. This leads to a contradiction.
Therefore Φ : `∞ (J) → B(H) is not completely bounded, and thus {uj , yj } it can
not be re-scaled to obtain a framing that admits a Hilbert space dilation.
Next we examine Example 3.9 of [CHL], which gave the first apparently nontrivial example of a bounded framing for a Hilbert space which is not a dual pair
of frames. A natural, geometric, Banach space dilation of it was constructed in
[CHL], along with a proof that it did not have any Hilbert dilation space. Both
the construction and the proof was nontrivial, and both were sketched out in [CHL]
without providing great detail. It was thought for a long time that this example
might be a key to this subject. But when we were finalizing this paper we found
a proof that this example can, in fact, be rescaled to give a dual pair of frames,
in fact a pair which consists of two copies of a single Parseval frame. Since this
example was a guiding example for a lengthy time, we include our analysis of it.
After obtaining this result, since we knew we still needed a true limiting example
for this theory, we managed to obtained the example in Theorem 5.1 above. The
[CHL] example points out how extremely difficult it can be in general to determine
when a given framing is scalable to a dual frame pair.
Theorem 5.4. Example 3.9 in [CHL] can be re-scaled to a dual pair of frames.
Proof. This result is very technical and it is a stand-alone result for the
present paper because the details of the proof are not used anywhere else in this
paper. So we must refer the reader to [CHL] for some of the terminology and
Banach space background. With this in hand this proof can be handily worked
through.
n
Fix any 1 < p < ∞ and p 6= 2 and natural number n. Let {eni }2i=1 be a
n
n
n
unconditional unit basis of `2p and {(eni )∗ }2i=1 be the dual of {eni }2i=1 . We denote
n
the Rademacher vectors in `2p by {rin }ni=1 , where
n
rin
=
2
1 X
2n/p
ij enj
j=1
with krin k`2pn = 1 and ij satisfying the condition
n
2
X
ij kj = δik · 2n .
j=1
Then {rin }ni=1 are linearly independent because they are the Rademacher
vectors.
Pn
n
So Wn = span{rin }ni=1 is a n-dimensional subspace of `2p . For any i=1 ai rin ∈ Wn ,
74
5. EXAMPLES
we have
n
X
ai rin i=1
X
2n
n ai X
n
ij ej n/p
i=1 2
2n
j=1
`p
n
! X
2
n
X
1 n
ai ij ej 2n/p j=1 i=1
=
`2pn
=
`2pn
n
p 1/p
2n
1 X X

.
a
i ij 2n/p j=1 i=1

(5.1)
=
Let sign(sin 2i πt), i = 0, 1, · · · , n be the Rademacher functions on [0, 1]. By
Theorem 2.b.3 in [LT], we have
p
p
n
Z 1 X
n
2n Z jn X
X
2
ai sign(sin 2i πt) dt =
ai sign(sin 2i πt) dt
j−1 0
2n
i=1
j=1
i=1
p
2n Z jn X
n
X
2
=
ai ij dt
j−1 n
2
j=1
i=1
n
p
n
2
X
1 X
a
=
.
i
ij
2n j=1
i=1
Let Ap , Bp be the constants as in Theorem 2.b.3 in [LT]. Then for any {ai }ni=1 , we
have
n
p 1/p
!1/2  2n
!1/2
n
n
X
X
X 1 X
2
2
ai ij 
≤ Bp
|ai |
,
≤
Ap
|ai |
2n i=1
i=1
j=1
i=1
that is
Ap
n
X
!1/2
2
|ai |
i=1
 n p 1/p
2 X
n
X
1
≤ n/p 
ai ij 
≤ Bp
2
j=1 i=1
Thus from equation (5.1), we obtain
!1/2 n
n
X
X
2
n
Ap
|ai |
≤
ai ri i=1
≤ Bp
`2pn
i=1
n
X
n
X
!1/2
|ai |
2
i=1
The induced Hilbert space norm is
n
X
n
ai ri i=1
i=1
k=1
Hn
=
n
X
i=1
!1/2
|ai |
2
,
|ai |
i=1
Let Hn be the span of {r1n , · · · , rnn } with the inner product
* n
+
n
n
X
X
X
n
n
ai ri ,
bk rk =
ai bi .
i=1
!1/2
2
.
.
5. EXAMPLES
75
and {r1n , · · · , rnn } is an orthonormal basis for Hn . Then Hn is a Hilbert space
isometrically isomorphism to l2n . The spaces Hn and Wn are naturally isomorphism as Banach space, being the same vector space with two different norms. Let
Un : Wn → Hn , Un x = x be this isomorphism.
n
Define a mapping Pn : `2p → Wn by the dual Rademachers
n
(rin )∗ =
2
1 X
2n/q
where 1/p + 1/q = 1. That is Pn (x) =
Pn
Pn (enk ) = i=1 (rin )∗ (enk )rin . Since
ij (enj )∗ ,
j=1
n
Pn
n ∗
n
i=1 (ri ) (x)ri
for any x ∈ `2p . So we have
n
2
1 X
(rin )∗ (enk ) =
2n/q
ij (enj )∗ (enk ) =
j=1
1
ik ,
2n/q
we get that
Pn (enk ) =
(5.2)
n
n
X
1
1 X
n
ik rin
r
=
ik
i
n/q
n/q
2
2
i=1
i=1
and
Pn2 (enk )
n
1 X
= Pn
2n/q
!
ik rin
i=1
P2n
=
n
1 X
2n/q
ik Pn (rin ) .
i=1
n
= δik · 2 , we have

2n
2n
X
1 X
1
ij enj  = n/p
ij Pn enj
Pn (rin ) = Pn  n/p
2
2
j=1
j=1
!
n
2
n
2n n
1 X
1 X
1 XX
n
=
ij
kj rk = n
ij kj rkn
2 j=1
2n/p j=1
2n/q k=1
k=1
 n

n
2
n
X
1 X X
 rkn = 1
=
δik · 2n · rkn
ij
kj
2n
2n
j=1
On the other hand, since
j=1 ij kj

k=1
k=1
= rin .
So we get
Pn2 (enk )
=
=
=
n
1 X
2n/q
i=1
n
1 X
2n/q
ik Pn (rin )
ik rin
i=1
n
Pn (ek ).
n
`2p
So Pn is a projection from
onto Wn and Pn (rin ) = rin .
To complete the proof, we need the following:
Lemma 5.5. The projections Pn are uniformly bounded in norm.
76
5. EXAMPLES
This lemma can be deduced implicitly from standard results in the literature
( c.f. [DJT, R, FHHMPZ] ). However we have not found it stated explicitly in
any references. Thus for self completeness we include the following proof which was
kindly shown to us by P. Casazza. It is short and self-contained but doesn’t give
the best uniform bound.
In order to prove Lemma 5.5, we need the following results from Lindenstrauss
and Tzafriri:
Lemma 5.6. There are constants Ap , Bp so that if {ri }m
i=1 are the Rademacher
vectors in `np , then for all scalars {ai }m
i=1 , we have
!1/2 m
!1/2
m
m
X
X
X
2
2
ai ri ≤ Bp
|ai |
.
Ap
|ai |
≤
n
i=1
i=1
i=1
`p
n
n
Lemma 5.7. If {ri }m
i=1 are the Rademacher vectors in `p , then for all x ∈ `q ,
!1/2
m
X
|ri (x)|2
≤ Bp kxk.
i=1
Pm
Pm
2
2 1/2
Proof. Give x ∈ `nq , choose {ai }m
=
i=1 so that
i=1 |ai | = 1 and
i=1 |ri (x)|
a
r
(x).
Then
we
get
i
i
i=1
!1/2
!
m
m
m
X
X
X
2
|ri (x)|
=
ai ri (x) =
ai ri (x)
Pm
i=1
i=1
≤
i=1
m
X
ai ri kxk ≤ Bp
i=1
m
X
!1/2
2
|ai |
kxk = Bp kxk.
i=1
Proof of Lemma 5.5: Let {ri }ni=1 (respectively, {ri∗ }ni=1 ) be the Rademachers
n
in
(respectively, `2q ) with 1/p + 1/q = 1. Then
n
`2p
ri∗ (rj ) = δi,j .
Now we check the norm of Pn . Now, applying our Lemma 5.7 twice,
kPn (x)k
=
sup
|f (Pn (x))|
kf k`2n =1
q
m
X
∗
=
sup ri (x)f (ri )
kf k 2n =1 `q
≤
i=1
m
X
sup
kf k`2n =1
i=1
q
≤
sup
!1/2
|ri∗ (x)|2
Bq kxkBp kf k
kf k`2n =1
q
= Bp Bq kxk.
We complete the proof of Lemma 5.5.
m
X
i=1
!1/2
2
|f (ri )|
5. EXAMPLES
77
Now we resume the proof of Theorem 5.4: Let Z and W be the Banach spaces
defined by
∞
X
n
1
n
Z=
⊕2 `2p = `2p ⊕2 · · · ⊕2 `2p ⊕2 · · ·
n=1
and
W =
∞
X
⊕2 Wn = W1 ⊕2 · · · ⊕2 Wn ⊕2 · · · ,
n=1
and let P be the projection from Z onto W defined by
P =
∞
X
⊕ 2 Pn = P1 ⊕ 2 · · · ⊕ 2 Pn ⊕ 2 · · · .
n=1
n
n
Since {eni }2i=1 is the unconditional unit basis of `2p , we obtain that
e1i ⊕2 0 ⊕2 · · · ⊕2 0 ⊕2 · · · ,
i = 1, 2
0 ⊕2 e2i ⊕2 · · · ⊕2 0 ⊕2 · · · ,
i = 1, 2, 3, 4
······
0 ⊕2 · · · ⊕2 eni ⊕2 0 ⊕2 · · · ,
i = 1, 2, · · · , 2n
······
is an uncondition basis of Z, and
(e1i )∗ ⊕2 0 ⊕2 · · · ⊕2 0 ⊕2 · · · ,
0
⊕2 (e2i )∗
⊕2 · · · ⊕2 0 ⊕2 · · · ,
i = 1, 2
i = 1, 2, 3, 4
······
0 ⊕2 · · · ⊕2 (eni )∗ ⊕2 0 ⊕2 · · · ,
i = 1, 2, · · · , 2n
······
is the dual of this basis. Thus
n
o
P (0 ⊕2 · · · ⊕2 eni ⊕2 0 ⊕2 · · · ) , P ∗ (0 ⊕2 · · · ⊕2 (eni )∗ ⊕2 0 ⊕2 · · · )
i=1,··· ,2n ,n=1,2,···
is a framing of W , where
P∗ =
∞
X
⊕2 Pn∗ = P1∗ ⊕2 · · · ⊕2 Pn∗ ⊕2 · · · ,
n=1
and
P ∗ (0 ⊕2 · · · ⊕2 (eni )∗ ⊕2 0 ⊕2 · · · ) = 0 ⊕2 · · · ⊕2 P ∗ (eni )∗ ⊕2 0 ⊕2 · · · .
Since each
Un : Wn → Hn ,
Un x = x
is an isomorphic operator, it follows that
U=
∞
X
⊕2 Un = U1 ⊕2 · · · ⊕2 Un ⊕2 · · ·
n=1
is an isomorphic operator. This implies that
n
o
U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · ) , (U −1 )∗ (0 ⊕2 · · · ⊕2 Pn∗ (eni )∗ ⊕2 0 ⊕2 · · · )
i=1,··· ,2n ,n=1,2,···
78
5. EXAMPLES
is a framing of the Hilbert space H, where
H=
∞
X
⊕2 H n = H 1 ⊕2 · · · ⊕ 2 H n ⊕2 · · · .
n=1
Now we prove that this framing has a Hilbert space dilation.
Take αin = 2n(1/q−1/2) ,P
i = 1, · · · , 2n , n = 1, 2, · · · . For any h = h1 ⊕2 · · · ⊕2
n
hn ⊕2 · · · ∈ H, where hn = k=1 ank rkn , n = 1, 2, · · · ,
hh, αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · )iH
=
2n(1/q−1/2) hh1 ⊕2 · · · ⊕2 hn ⊕2 · · · , 0 ⊕2 · · · ⊕2 Un (Pn (eni )) ⊕2 0 ⊕2 · · · iH
=
2n(1/q−1/2) hhn , Un (Pn (eni ))iHn
2n(1/q−1/2) hhn , Pn (eni )iHn .
Pn
1
n
From Pn (eni ) = 2n/q
j=1 ji rj , we get that
+
* n
n
X
1 X
n
n n
n
ji rj
ak rk , n/q
hhn , Pn (ei )iHn =
2
j=1
k=1
=
=
Hn
n
1 X
2n/q
ank ki .
k=1
Thus
=
(5.3)
=
hh, αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · )iH
n
1 X n
2n(1/q−1/2) · n/q
ak ki
2
k=1
2−n/2
n
X
ank ki .
k=1
On the other hand,
1
−1 ∗
∗ n ∗
h, n (U ) (0 ⊕2 · · · ⊕2 Pn (ei ) ⊕2 0 ⊕2 · · · )
αi
H
n(1/2−1/q)
= 2
h, 0 ⊕2 · · · ⊕2 (Un−1 )∗ (Pn∗ (eni )∗ ) ⊕2 0 ⊕2 · · · H
= 2n(1/2−1/q) hn , (Un−1 )∗ (Pn∗ (eni )∗ ) H
n
= 2n(1/2−1/q) Pn∗ (eni )∗ Un−1 hn
=
2n(1/2−1/q) Pn∗ (eni )∗ (hn )
=
2n(1/2−1/q) (eni )∗ (Pn (hn ))
2n(1/2−1/q) (eni )∗ (hn )
!
n
X
n(1/2−1/q) n ∗
n n
= 2
(ei )
ak rk
=
k=1
=
2n(1/2−1/q)
n
X
ank (eni )∗ (rkn ) .
k=1
Since

2n
2n
X
1 X
1
1
n
n ∗
n
n ∗
kj ej = n/p
kj (eni )∗ enj = n/p ki ,
(ei ) (rk ) = (ei )
n/p
2
2
2
j=1
j=1

5. EXAMPLES
79
we have
=
=
1
−1 ∗
∗ n ∗
h, n (U ) (0 ⊕2 · · · ⊕2 Pn (ei ) ⊕2 0 ⊕2 · · · )
αi
H
n
X
1
2n(1/2−1/q)
ank n/p ki
2
k=1
2
n(1/2−1/q−1/p)
n
X
ank ki
k=1
=
(5.4)
2−n/2
n
X
ank ki .
k=1
From (5.3) and (5.3), we know that
αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · ) =
1
(U −1 )∗ (0 ⊕2 · · · ⊕2 Pn∗ (eni )∗ ⊕2 0 ⊕2 · · · ) .
αni
Now we show that
n
o
αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · )
i=1,··· ,2n ,n=1,2,···
is the Parseval frame of H.
Pn
For any h = h1 ⊕2 · · · ⊕2 hn ⊕2 · · · ∈ H, where hn = k=1 ank rkn , n = 1, 2, · · · ,
we have
n
∞
∞ X
X
X
|ank |2 .
khk2 =
khn k2 =
n=1
From
P2n
j=1 ij kj
n=1 k=1
= δik · 2n , for any {ank }k=1,··· ,n , we have
2
2n
n
n
X
X
1 X n a
=
|ank |2 .
k ki 2n i=1
k=1
k=1
Hence we get
n
2
∞ X
X
2
|hh, αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · )iH |
n=1 i=1
=
=
=
2
∞ X
2n n
X
1 X n ak ki n/2
2
n=1 i=1
k=1
2
∞ X
2n
n
X
1 X n ak ki 2n n=1 i=1
∞ X
n
X
k=1
|ank |2
n=1 k=1
=
khk2 .
Therefore
n
o
αin U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · )
i=1,··· ,2n ,n=1,2,···
80
5. EXAMPLES
is the Parseval frame of H. Finally by Theorem 3.8, we obtain that
n
o
U (0 ⊕2 · · · ⊕2 Pn (eni ) ⊕2 0 ⊕2 · · · ) , (U −1 )∗ (0 ⊕2 · · · ⊕2 Pn∗ (eni )∗ ⊕2 0 ⊕2 · · · )
i=1,··· ,2n ,n=1,2,···
has a Hilbert space dilation. This completes the proof of Theorem 5.4.
Bibliography
[Arv] W. Arveson, Dilation theory yesterday and today, A glimpse at Hilbert space operators,
99–123, Oper. Theory Adv. Appl., 207, Birkhäuser Verlag, Basel, 2010.
[Bun] J. Bunce, The similarity problem for representations of C ∗ -algebras, Proc. Amer. Math.
Soc., 81(1981), 409–413.
[BW] L. J. Bunce and J. D. Maitland Wright, The Mackey-Gleason problem for vector measures
on projections in a von Neumann algebra, J. London Math. Soc., 49 (1994), 133–149.
[CDOSZ] P. G. Casazza, S. J. Dilworth, E. Odell, Th. Schlumprecht, and A. Zsak, Coefficient
Quantization for Frames in Banach Spaces, J. Math. Anal. Appl. 348 (2008), 66–86.
[CHL] P. G. Casazza, D. Han, and D. R. Larson, Frames for Banach spaces, The functional and
harmonic analysis of wavelets and frames (San Antonio, TX, 1999), Contemp. Math. 247
(1999), 149–182.
[CKL] P. G. Casazza, G. Kutyniok, and S. Li, Fusion frames and distributed processing, Appl.
Comput. Harmon. Anal. 25 (2008), 114–132.
[Cho] M. D. Choi, Completely Positive Linear Maps on Complex matrices, Lin. Alg. Appl., 12
(1975), 285-290.
[Chr] E. Christensen, On non-selfadjoint representations of operator algebras, Amer. J. Math.,
103(1981), 817–834.
[Ch] O. Christensen, An introduction to frames and Riesz bases, Birkhäuser, 2003.
[D] I.Daubechies, Ten Lectures on Wavelets, SIAM Philadelphia, 1992.
[DJT] J. Diestel, H. Jarchow and A. Tonge, Absolutely Summing Operators, Cambridge University
Press, Cambridge, 1995.
[DL] X. Dai and D. R. Larson, Wandering vectors for unitary systems and orthogonal wavelets,
Mem. Amer. Math. Soc. 134, 1998.
[DS] R.J. Duffin and A.C. Schaeffer, A class of nonharmonic Fourier series, Trans. Amer. Math.
Soc., 72 (1952), 341–366.
[FHHMPZ] M. Fabian, P. Habala, P. Hájek, V. Montesinos, J. Pelant and V. Zizler, Functional
Analysis and Infinite Dimensional Geometry, CMS Books in Mathematics, Springer Verlag
2001.
[FR] M. Fornasier and H. Rauhut, Continuous frames, function spaces, and the discretization
problem, J. Fourier Anal. Appl. 11 (2005) 245–287.
[GH] J.-P. Gabardo and D. Han. Frames associated with measurable spaces, Adv. Comput. Math.
18 (2003), 127–147.
[GH2] J.-P. Gabardo and D. Han. Frame representations for group-like unitary operator systems,
J. Operator Theory , 49 (2003), 223–244.
[Haa] U. Haagerup, Solutions of the similarity problem for cyclic representations of C ∗ -algebras,
Ann. Math., 118 (1983), 215–240.
[HA] D. W. Hadwin, Dilations and Hahn decompositions for linear maps, Canad. J. Math. 33
(1981), 826–839.
[Han1] D. Han Frame representations and parseval duals with applications to Gabor frames,
Trans. Amer. Math. Soc., 360 (2008), 3307–3326.
[Han2] D. Han Dilations and completions for Gabor systems, J. Fourier Anal. Appl., 15 (2009),
201–217.
[HL] D. Han and D. R. Larson, Frames, bases and group representations, Mem. Amer. Math.
Soc. 697, 2000.
[K1] R. V. Kadison, On the orthogonalization of operator representations, Amer. J. Math., 77
(1955), 600–620.
81
82
BIBLIOGRAPHY
[KLZ] V. Kaftal, D. R. Larson, and S. Zhang, Operator valued frames, Trans. Amer. Math. Soc.,
361 (2009),6349–6385.
[LO1] S. Li and H. Ogawa, Pseudoframes for subspaces with applications, J. Fourier Anal. Appl.10
(2004), 409–431.
[LO2] S. Li, and H. Ogawa, Pseudo-duals of frames with applications, Appl. Comput. Harmon.
Anal. 11(2) (2001) 289–304.
[LT] J. Lindenstrauss and L. Tzafriri, Classical Banach spaces. I. Sequence spaces, SpringerVerlag, Berlin, 1979.
[M] G. W. Mackey, Imprimitivity for representations of locally compact groups, I. Proc. Nat.
Acad. Sci. U.S.A. 35 (1949), 537–545.
[Os] H. Osaka, Completely bounded maps between the preduals of von Neumann algebras, Proc.
Amer. Math. Soc., 111 (4) (1991), 961–965.
[Pa] V. Paulsen, Completely bounded maps and operator algebras, Cambridge University Press,
2002.
[Pi] G. Pisier, Simliarity problems and completely bounded maps, 2nd, expanded edition, SpringerVerlag lecture Notes, Vol. 1618, 2001.
[R] R. Ryan, Introduction to Tensor Products of Banach Spaces, Springer-Verlag, London, 2003.
[Rud] W.Rudin, Real and Complex Analysis, McGraw-Hill,Inc., 1997.
[Si] I. Singer, Bases in Banach Spaces I, Springer-Verlag (1970).
[St] W. F. Stinespring, Positive Functions on C*-algebras, Proc. Amer. Math. Soc., 6 (1955),
211–216.
[Su] W. Sun, G-frames and g-Riesz bases, J. Math. Anal. Appl. 232 (2006), 437–452.